Node Reference

Auto-generated from Schema() metadata. 85 nodes registered.

Node	Description	Params
a2a_agent	Sends the input as one task to an A2A agent (message/send with tasks/send fallback), polls tasks/get until a terminal...	4
agent	Autonomous agent node with ReAct reasoning loop and tool use capabilities	9
anthropic	Call Anthropic Claude via an OpenAI-compatible proxy (LiteLLM/one-api). Native api.anthropic.com is NOT OpenAI-compat...	15
ark	Call Volcengine Ark LLM API	15
ascend	Call Ascend LLM API	15
baichuan	Call Baichuan LLM API	15
call	Call another workflow file	2
cambricon	Call Cambricon MLU LLM API	15
cerebras	Call Cerebras LLM API	15
clarify	Pre-execution ambiguity checker: identifies unclear requirements and generates clarifying questions (ACQUIRE framework)	8
cli_agent	Runs one bounded task via an external CLI agent subprocess (codex exec, claude -p, gemini -p, or a generic command) w...	9
cli_session	交互式CLI会话节点。支持上下文保持、命令历史、快捷键、流式输出和自动补全，提供类...	5
code_interpreter	Execute Python/Node.js/Rust code in a sandboxed environment with file I/O	6
code_knowledge_graph	Semantic code knowledge graph with vector retrieval, 158 language support, MCP tool exposure, and token-efficient rev...	13
code_review	Hybrid code review combining deterministic rule engine (NPE, thread-safety, security) with LLM deep analysis. Inspire...	11
codex_agent	Runs one bounded Codex agent task via codex exec and returns its final output. Requires the codex CLI (https://gith...	7
combine	Combine multiple inputs into one	1
compress	Intelligent context compression with 6 algorithms: extractive, keyword, cluster, sliding_window, hybrid (headroom-ins...	7
condition	Evaluate conditional expressions (contains, equals, regex, empty)	2
coze	Call Coze LLM API	15
critic	Critic agent that reviews output, identifies issues, and suggests improvements	8
deepseek	Call DeepSeek LLM API	15
doc_gen	AI自动文档生成节点。自动生成和更新代码库文档，支持README、API文档、函数注释、模块...	6
doc_parse	Parse documents (PDF/images/HTML) into text, LaTeX, or HTML table format	7
drone	Control MAVLink-compatible drones (PX4/ArduPilot) via HTTP bridge. Supports arm/disarm, takeoff, land, RTL, waypoint ...	17
email_send	Send an email over SMTP (implicit TLS on port 465, STARTTLS elsewhere; plaintext only for loopback relays)	12
engineer_skills	预置工程技能包，覆盖前端/后端/DevOps/架构/安全/数据/信创七大领域共 24 项技能。支持...	5
evaluator	Evaluator agent that scores output against criteria with structured rubrics	9
execute	Execute shell commands (disabled in safe mode)	3
fastgpt	Call FastGPT API	5
fetch_url	Fetch content from a URL	3
file_read	Read content from a file. Automatically redacts secrets (API keys, tokens, .env files) by default for privacy — set...	3
file_watch	Polls a file or directory for create/modify/delete events and returns them as JSON. Suitable for log-monitor and file...	6
file_write	Write content to a file	4
files_list	List files under a files/notes connector root (relative paths + sizes). Skips dotfiles/dot-directories and symlinks. ...	3
fireworks	Call Fireworks AI LLM API	15
gemini	Call Google Gemini LLM API	15
glm	Call GLM LLM API	15
groq	Call Groq LLM API	15
http_request	Make HTTP requests with custom method, headers, and body	13
human_in_loop	Human approval gate — pauses workflow for human review and approval before continuing	5
hunyuan	Call Tencent Hunyuan LLM API	15
hygon	Call Hygon DCU LLM API	15
ima	Call IMA Copilot LLM API	15
internlm	Call InternLM LLM API	15
json_parse	Parse and extract JSON data	1
kimi	Call Kimi LLM API	15
knowledge_graph	Knowledge graph node - extract entities/relations, build graph, query and traverse	13
llm_router	Smart LLM router that automatically selects the best provider with fallback, quota tracking, and cost optimization	5
memory	AI Agent memory infrastructure with session-isolated persistent knowledge graph engine. Supports multi-session parall...	16
mimo	Call MiMo LLM API	15
minimax	Call MiniMax LLM API	15
mistral	Call Mistral LLM API	15
multimodal	Multimodal node for image analysis, OCR, and audio transcription using vision-capable LLMs	10
notify	Send notifications (stdout, stderr, slack, discord, telegram, feishu, dingtalk, wecom, webhook)	10
nvidia	Call NVIDIA NIM LLM API	15
office	Read .docx/.xlsx/.pptx documents (text, tables, slides) using pure-Go OOXML parsing	5
ollama	Call Ollama local LLM server	3
openai	Call OpenAI LLM API	15
openrouter	Call OpenRouter LLM API	15
perplexity	Call Perplexity LLM API	15
pipeline	Dependency-based parallel workflow executor: steps run as soon as their dependencies are met, no global barriers (Tun...	2
planner	Task decomposition agent that breaks complex goals into actionable steps	9
preference	User preference memory: store, retrieve, and learn user habits across sessions (MemSlides-inspired user profiling)	7
qianfan	Call Baidu Qianfan LLM API	15
qwen	Call Qwen LLM API	15
rag	Retrieval Augmented Generation node - chunk documents, search by query, and assemble context	7
reflector	Self-reflection agent that critiques output and iteratively improves it (Reflexion pattern)	7
researcher	Research agent that fetches information from URLs and summarizes findings	7
search_aggregate	Multi-platform search aggregator with real-signal ranking: Reddit/Twitter/YouTube/HN/GitHub, sorted by votes/comments...	8
session_manager	Multi-session memory management. Create isolated sessions, fork a session from a parent, merge sessions, and share fa...	11
siliconflow	Call SiliconFlow LLM API	15
skill_distill	Distill methodologies from books, videos, podcasts, and documents into callable skills. Supports workflow, decision, ...	6
sql_query	Execute SQL via database/sql. The driver must be registered by the host program. Uses parameterized queries (? or $1)...	9
stepfun	Call StepFun LLM API	15
structured_output	LLM-driven structured output with local JSON Schema validation and self-correction retries	11
supervisor	Advanced supervisor with MoE routing, MindSearch deep research, 232+ domain specialists, and collaboration templates....	16
template_render	Render Go templates with input data	2
together	Call Together AI LLM API	15
transform	Transform text using string operations	1
verify	Agent-as-a-Judge verifier that validates outputs, claims, and results against specified criteria	10
wait	Pause the workflow for a duration (delay/sleep), then pass input through	1
xai	Call xAI Grok LLM API	15
xverse	Call XVERSE LLM API	15
yi	Call Yi LLM API	15

---

a2a_agent

Sends the input as one task to an A2A agent (message/send with tasks/send fallback), polls tasks/get until a terminal state and returns the artifacts/status text. Consecutive failures trip a per-agent circuit breaker: calls fail fast with a circuit-open error until an agent-card probe succeeds, so one dead endpoint cannot burn the full timeout on every step.

• Input: string - the task/prompt for the remote agent
• Output: string - the agent's artifacts/status message text

Parameters

Name	Type	Required	Default	Description
agent	string	No		Registered A2A agent name (see aflare agent list)
url	string	No		A2A service endpoint (http/https); required when no agent name is given
api_key_env	string	No		Environment variable holding the bearer token for the remote agent
timeout	string	No		Overall step timeout, e.g. 30s, 10m, 1h (default 10m, max 60m)

---

agent

Autonomous agent node with ReAct reasoning loop and tool use capabilities

• Input: string - the task or question for the agent
• Output: string - the agent's final answer

Parameters

Name	Type	Required	Default	Description
provider	string	No	ollama	LLM provider: ollama, openai, deepseek, glm, kimi, qwen, mistral, yi (default: ollama)
model	string	No	llama3	Model name (default: llama3)
api_key	string	No		API key (for cloud providers)
endpoint	string	No		API endpoint URL
system	string	No		System prompt / role definition for the agent
tools	string	No	fetch_url,json_parse	Comma-separated list of tools to enable: fetch_url,http_request,file_read,file_write,json_parse,transform,combine,ollama,openai,code_interpreter,execute
max_iterations	string	No	10	Maximum number of ReAct iterations (default: 10)
enable_thinking	string	No	false	Enable deep thinking / chain-of-thought mode (default: false)
show_thinking	string	No	true	Show the thinking chain in output (default: true)

---

anthropic

Call Anthropic Claude via an OpenAI-compatible proxy (LiteLLM/one-api). Native api.anthropic.com is NOT OpenAI-compatible and will 404; configure endpoint to point at a proxy that translates the protocol.

• Input: string - user message content
• Output: string - AI response content

Parameters

Name	Type	Required	Default	Description
model	string	No	claude-sonnet-5	Model name (default: claude-sonnet-5)
api_key	string	No		Anthropic API key (or set ANTHROPIC_API_KEY env var)
endpoint	string	No	https://api.anthropic.com/v1	API base URL (default: https://api.anthropic.com/v1)
system	string	No		System prompt
temperature	string	No		Sampling temperature 0.0-2.0 (default: provider default)
max_tokens	string	No		Max tokens to generate
top_p	string	No		Nucleus sampling probability mass 0.0-1.0
frequency_penalty	string	No		Penalty for repeated tokens -2.0 to 2.0
presence_penalty	string	No		Penalty for new tokens -2.0 to 2.0
stop	string	No		Stop sequences (comma-separated, e.g. '\n,END')
seed	string	No		Random seed for deterministic sampling (int)
response_format	string	No		Structured output: 'json_object' or 'json_schema:<schema_json>'
tools	string	No		JSON array of tool definitions for function calling
tool_choice	string	No		Tool selection: 'none', 'auto', or JSON object
user	string	No		End-user identifier for provider-side abuse monitoring

---

ark

Call Volcengine Ark LLM API

• Input: string - user message content
• Output: string - AI response content

Parameters

Name	Type	Required	Default	Description
model	string	No	doubao-seed-2-1-pro-260628	Model name (default: doubao-seed-2-1-pro-260628)
api_key	string	No		Volcengine Ark API key (or set ARK_API_KEY env var)
endpoint	string	No	https://ark.cn-beijing.volces.com/api/v3	API base URL (default: https://ark.cn-beijing.volces.com/api/v3)
system	string	No		System prompt
temperature	string	No		Sampling temperature 0.0-2.0 (default: provider default)
max_tokens	string	No		Max tokens to generate
top_p	string	No		Nucleus sampling probability mass 0.0-1.0
frequency_penalty	string	No		Penalty for repeated tokens -2.0 to 2.0
presence_penalty	string	No		Penalty for new tokens -2.0 to 2.0
stop	string	No		Stop sequences (comma-separated, e.g. '\n,END')
seed	string	No		Random seed for deterministic sampling (int)
response_format	string	No		Structured output: 'json_object' or 'json_schema:<schema_json>'
tools	string	No		JSON array of tool definitions for function calling
tool_choice	string	No		Tool selection: 'none', 'auto', or JSON object
user	string	No		End-user identifier for provider-side abuse monitoring

---

ascend

Call Ascend LLM API

• Input: string - user message content
• Output: string - AI response content

Parameters

Name	Type	Required	Default	Description
model	string	No	qwen2.5-7b	Model name (default: qwen2.5-7b)
api_key	string	No		Ascend API key (or set ASCEND_API_KEY env var)
endpoint	string	No	http://localhost:8080/v1	API base URL (default: http://localhost:8080/v1)
system	string	No		System prompt
temperature	string	No		Sampling temperature 0.0-2.0 (default: provider default)
max_tokens	string	No		Max tokens to generate
top_p	string	No		Nucleus sampling probability mass 0.0-1.0
frequency_penalty	string	No		Penalty for repeated tokens -2.0 to 2.0
presence_penalty	string	No		Penalty for new tokens -2.0 to 2.0
stop	string	No		Stop sequences (comma-separated, e.g. '\n,END')
seed	string	No		Random seed for deterministic sampling (int)
response_format	string	No		Structured output: 'json_object' or 'json_schema:<schema_json>'
tools	string	No		JSON array of tool definitions for function calling
tool_choice	string	No		Tool selection: 'none', 'auto', or JSON object
user	string	No		End-user identifier for provider-side abuse monitoring

---

baichuan

Call Baichuan LLM API

• Input: string - user message content
• Output: string - AI response content

Parameters

Name	Type	Required	Default	Description
model	string	No	Baichuan4	Model name (default: Baichuan4)
api_key	string	No		Baichuan API key (or set BAICHUAN_API_KEY env var)
endpoint	string	No	https://api.baichuan-ai.com/v1	API base URL (default: https://api.baichuan-ai.com/v1)
system	string	No		System prompt
temperature	string	No		Sampling temperature 0.0-2.0 (default: provider default)
max_tokens	string	No		Max tokens to generate
top_p	string	No		Nucleus sampling probability mass 0.0-1.0
frequency_penalty	string	No		Penalty for repeated tokens -2.0 to 2.0
presence_penalty	string	No		Penalty for new tokens -2.0 to 2.0
stop	string	No		Stop sequences (comma-separated, e.g. '\n,END')
seed	string	No		Random seed for deterministic sampling (int)
response_format	string	No		Structured output: 'json_object' or 'json_schema:<schema_json>'
tools	string	No		JSON array of tool definitions for function calling
tool_choice	string	No		Tool selection: 'none', 'auto', or JSON object
user	string	No		End-user identifier for provider-side abuse monitoring

---

call

Call another workflow file

• Input: string - input data to pass to the called workflow
• Output: string - output from the called workflow

Parameters

Name	Type	Required	Default	Description
workflow	string	Yes		Path to the workflow file to call
vars	string	No		JSON or key=value pairs to pass as workflow variables (e.g. topic=AI,model=gpt-4)

---

cambricon

Call Cambricon MLU LLM API

• Input: string - user message content
• Output: string - AI response content

Parameters

Name	Type	Required	Default	Description
model	string	No	qwen2.5-7b	Model name (default: qwen2.5-7b)
api_key	string	No		Cambricon MLU API key (or set CAMBRICON_API_KEY env var)
endpoint	string	No	http://localhost:8081/v1	API base URL (default: http://localhost:8081/v1)
system	string	No		System prompt
temperature	string	No		Sampling temperature 0.0-2.0 (default: provider default)
max_tokens	string	No		Max tokens to generate
top_p	string	No		Nucleus sampling probability mass 0.0-1.0
frequency_penalty	string	No		Penalty for repeated tokens -2.0 to 2.0
presence_penalty	string	No		Penalty for new tokens -2.0 to 2.0
stop	string	No		Stop sequences (comma-separated, e.g. '\n,END')
seed	string	No		Random seed for deterministic sampling (int)
response_format	string	No		Structured output: 'json_object' or 'json_schema:<schema_json>'
tools	string	No		JSON array of tool definitions for function calling
tool_choice	string	No		Tool selection: 'none', 'auto', or JSON object
user	string	No		End-user identifier for provider-side abuse monitoring

---

cerebras

Call Cerebras LLM API

• Input: string - user message content
• Output: string - AI response content

Parameters

Name	Type	Required	Default	Description
model	string	No	llama-3.3-70b	Model name (default: llama-3.3-70b)
api_key	string	No		Cerebras API key (or set CEREBRAS_API_KEY env var)
endpoint	string	No	https://api.cerebras.ai/v1	API base URL (default: https://api.cerebras.ai/v1)
system	string	No		System prompt
temperature	string	No		Sampling temperature 0.0-2.0 (default: provider default)
max_tokens	string	No		Max tokens to generate
top_p	string	No		Nucleus sampling probability mass 0.0-1.0
frequency_penalty	string	No		Penalty for repeated tokens -2.0 to 2.0
presence_penalty	string	No		Penalty for new tokens -2.0 to 2.0
stop	string	No		Stop sequences (comma-separated, e.g. '\n,END')
seed	string	No		Random seed for deterministic sampling (int)
response_format	string	No		Structured output: 'json_object' or 'json_schema:<schema_json>'
tools	string	No		JSON array of tool definitions for function calling
tool_choice	string	No		Tool selection: 'none', 'auto', or JSON object
user	string	No		End-user identifier for provider-side abuse monitoring

---

clarify

Pre-execution ambiguity checker: identifies unclear requirements and generates clarifying questions (ACQUIRE framework)

• Input: string - the task or goal to analyze for ambiguity
• Output: string - JSON with clarification result: needs_clarification, questions, confidence, clarified_goal

Parameters

Name	Type	Required	Default	Description
provider	string	No	ollama	LLM provider (default: ollama)
model	string	No	llama3	Model name (default: llama3)
api_key	string	No		API key
endpoint	string	No		API endpoint URL
threshold	string	No	70	Confidence threshold 0-100, below this trigger clarification (default: 70)
max_questions	string	No	5	Max clarification questions to ask (default: 5)
context	string	No		Additional context about the task
user_answers	string	No		JSON object of user's answers to previous questions (question -> answer)

---

cli_agent

Runs one bounded task via an external CLI agent subprocess (codex exec, claude -p, gemini -p, or a generic command) with timeout, sandbox and audit. Requires the agent CLI installed and authenticated. Consecutive failures trip a per-agent circuit breaker (fast circuit-open failures until the agent recovers).

• Input: string - the task/prompt for the agent
• Output: string - the agent's final answer (stdout)

Parameters

Name	Type	Required	Default	Description
agent	string	No		Registered agent name (see aflare agent list); overrides: binary/profile/model/sandbox can still be set per step
binary	string	No		Agent executable (bare name or absolute path); required when no agent name is given
profile	string	No		CLI profile: codex, claude, gemini, generic (default: generic when inline)
model	string	No		Model the agent should use (forwarded as a validated flag value)
sandbox	string	No		codex: strict, permissive, danger-full-access (default strict)
approval_policy	string	No		codex: never, on-failure, on-request, untrusted (default never); claude/generic: only never is supported
max_turns	string	No		Maximum agent turns, 0 for unlimited (default 0)
cwd	string	No		Working directory for the agent (must exist)
timeout	string	No		Overall step timeout, e.g. 30s, 10m, 1h (default 10m, max 60m)

---

cli_session

交互式CLI会话节点。支持上下文保持、命令历史、快捷键、流式输出和自动补全，提供类似Claude Code的流畅CLI体验。

• Input: string - 用户输入的命令或消息
• Output: string - JSON格式的会话响应

Parameters

Name	Type	Required	Default	Description
model	string	No	auto	使用的模型（默认auto，由路由层选择）
session_id	string	No		会话ID（自动生成或指定）
max_history	int	No	50	最大历史记录数（默认50）
streaming	bool	No	true	流式输出（默认true）
theme	string	No	dark	主题（light/dark，默认dark）

---

code_interpreter

Execute Python/Node.js/Rust code in a sandboxed environment with file I/O

• Input: string - stdin for the code (optional)
• Output: string - stdout, stderr, and generated files

Parameters

Name	Type	Required	Default	Description
code	string	Yes		Code to execute
language	string	No	python	Programming language: python, nodejs, rust (default: python)
timeout	string	No		Execution timeout (default based on security level)
work_dir	string	No		Working directory for code execution (default: temp dir)
save_outputs	string	No	true	If true, save output files to work_dir (default: true)
network	string	No	false	Allow network access during execution (L0/L1 only, default: false)

---

code_knowledge_graph

Semantic code knowledge graph with vector retrieval, 158 language support, MCP tool exposure, and token-efficient review. Supports incremental updates and persistent indexing.

• Input: string - optional query context or MCP tool call
• Output: string - JSON format with entities, relations, concepts, query results, or MCP tool response

Parameters

Name	Type	Required	Default	Description
path	string	Yes		Code path to analyze
mode	string	No	build_and_query	Mode: build/build_and_query/query_only/incremental/mcp_tool (default: build_and_query)
query	string	No		Query statement (text or vector)
query_type	string	No	semantic	Query type: semantic/symbol/path/relation (default: semantic)
top_k	int	No	10	Number of results to return (1-100, default: 10)
threshold	float	No	0.7	Similarity threshold 0.0-1.0 (default: 0.7)
vector_dim	int	No	384	Vector dimension (default: 384)
mcp_tool	string	No		MCP tool to call: list_entities/search_graph/analyze_dependencies/get_entity_details/list_relations/generate_summary
entity_name	string	No		Entity name for get_entity_details tool
token_efficient	bool	No	true	Enable token-efficient review mode (default: true)
incremental_update	bool	No	false	Use incremental update (only process changed files)
use_cache	bool	No	true	Use persistent cache index (default: true)
force_rebuild	bool	No	false	Force rebuild index from scratch (default: false)

---

code_review

Hybrid code review combining deterministic rule engine (NPE, thread-safety, security) with LLM deep analysis. Inspired by alibaba/open-code-review.

• Input: string - the code to review
• Output: string - structured code review with findings and suggestions

Parameters

Name	Type	Required	Default	Description
provider	string	No	ollama	LLM provider (default: ollama)
model	string	No	llama3	Model name (default: llama3)
api_key	string	No		API key
endpoint	string	No		API endpoint URL
language	string	No		Programming language (auto-detected if not specified)
focus	string	No	all	Review focus: all, bugs, security, style, performance (default: all)
severity	string	No	medium	Minimum severity: low, medium, high, critical (default: medium)
use_rules	string	No	true	Run deterministic rule engine before LLM (default: true)
use_llm	string	No	true	Run LLM deep analysis (default: true)
auto_clarify	string	No	false	Run ACQUIRE-style clarification before review (default: false)
clarify_threshold	string	No	70	Confidence threshold for auto-clarification 0-100 (default: 70)

---

codex_agent

Runs one bounded Codex agent task via codex exec and returns its final output. Requires the codex CLI (https://github.com/openai/codex) installed and authenticated.

• Input: string - the task/prompt for the Codex agent
• Output: string - the Codex agent's final answer (stdout)

Parameters

Name	Type	Required	Default	Description
binary	string	No	codex	Path to the codex executable (default: codex, resolved from PATH)
model	string	No		Model to use (passed as --model, e.g. gpt-5.6; empty = codex default)
sandbox	string	No	strict	Codex sandbox level: strict, permissive, danger-full-access (default: strict)
approval_policy	string	No	never	Approval policy for non-interactive runs: never, on-failure, on-request, untrusted (default: never)
max_turns	string	No	0	Maximum agent turns, 0 for unlimited (default: 0)
cwd	string	No		Working directory for the agent (must exist; default: current directory)
timeout	string	No	10m	Overall step timeout, e.g. 30s, 10m, 1h (default: 10m, max 60m)

---

combine

Combine multiple inputs into one

• Input: string - input text to format
• Output: string - formatted output

Parameters

Name	Type	Required	Default	Description
format	string	No	text	Output format: text, markdown, csv, json (default: text)

---

compress

Intelligent context compression with 6 algorithms: extractive, keyword, cluster, sliding_window, hybrid (headroom-inspired, 60-95% token reduction)

• Input: string - text to compress
• Output: string - compressed text with metadata

Parameters

Name	Type	Required	Default	Description
algorithm	string	No	hybrid	extract
ratio	string	No	0.2	Target compression ratio 0.01-1.0, lower=more aggressive (default: 0.2)
max_chars	string	No	4000	Maximum output characters (default: 4000)
preserve_headers	string	No	true	Preserve section headers (default: true)
preserve_numbers	string	No	true	Preserve sentences with numbers/stats (default: true)
output	string	No	text	text
keywords	string	No	0	Also extract top N keywords (default: 0)

---

condition

Evaluate conditional expressions (contains, equals, regex, empty)

• Input: string - the text to evaluate against
• Output: string - 'true' or 'false'

Parameters

Name	Type	Required	Default	Description
expr	string	Yes		Condition expression (e.g. contains:foo, equals:bar, regex:^test, empty, not_empty)
condition	string	No		Alias for expr

---

coze

Call Coze LLM API

• Input: string - user message content
• Output: string - AI response content

Parameters

Name	Type	Required	Default	Description
model	string	No		Model name (default: )
api_key	string	No		Coze API key (or set COZE_API_KEY env var)
endpoint	string	No	https://api.coze.cn/v1	API base URL (default: https://api.coze.cn/v1)
system	string	No		System prompt
temperature	string	No		Sampling temperature 0.0-2.0 (default: provider default)
max_tokens	string	No		Max tokens to generate
top_p	string	No		Nucleus sampling probability mass 0.0-1.0
frequency_penalty	string	No		Penalty for repeated tokens -2.0 to 2.0
presence_penalty	string	No		Penalty for new tokens -2.0 to 2.0
stop	string	No		Stop sequences (comma-separated, e.g. '\n,END')
seed	string	No		Random seed for deterministic sampling (int)
response_format	string	No		Structured output: 'json_object' or 'json_schema:<schema_json>'
tools	string	No		JSON array of tool definitions for function calling
tool_choice	string	No		Tool selection: 'none', 'auto', or JSON object
user	string	No		End-user identifier for provider-side abuse monitoring

---

critic

Critic agent that reviews output, identifies issues, and suggests improvements

• Input: string - the content to be reviewed and critiqued
• Output: string - structured critique with issues and improvement suggestions

Parameters

Name	Type	Required	Default	Description
provider	string	No	ollama	LLM provider (default: ollama)
model	string	No	llama3	Model name (default: llama3)
api_key	string	No		API key
endpoint	string	No		API endpoint URL
role	string	No	general	Critic role: general, code, writing, security, design (default: general)
criteria	string	No		Custom evaluation criteria (comma-separated)
output_format	string	No	markdown	Output format: markdown, json, bullet_points (default: markdown)
suggest_improvements	string	No	true	Whether to suggest improvements: true/false (default: true)

---

deepseek

Call DeepSeek LLM API

• Input: string - user message content
• Output: string - AI response content

Parameters

Name	Type	Required	Default	Description
model	string	No	deepseek-chat	Model name (default: deepseek-chat)
api_key	string	No		DeepSeek API key (or set DEEPSEEK_API_KEY env var)
endpoint	string	No	https://api.deepseek.com/v1	API base URL (default: https://api.deepseek.com/v1)
system	string	No		System prompt
temperature	string	No		Sampling temperature 0.0-2.0 (default: provider default)
max_tokens	string	No		Max tokens to generate
top_p	string	No		Nucleus sampling probability mass 0.0-1.0
frequency_penalty	string	No		Penalty for repeated tokens -2.0 to 2.0
presence_penalty	string	No		Penalty for new tokens -2.0 to 2.0
stop	string	No		Stop sequences (comma-separated, e.g. '\n,END')
seed	string	No		Random seed for deterministic sampling (int)
response_format	string	No		Structured output: 'json_object' or 'json_schema:<schema_json>'
tools	string	No		JSON array of tool definitions for function calling
tool_choice	string	No		Tool selection: 'none', 'auto', or JSON object
user	string	No		End-user identifier for provider-side abuse monitoring

---

doc_gen

AI自动文档生成节点。自动生成和更新代码库文档，支持README、API文档、函数注释、模块文档、更新日志、教程和架构文档等多种类型，让代码库对AI Agent更友好。

• Input: string - 代码内容或文档生成指令
• Output: string - 生成的文档内容（markdown或JSON格式）

Parameters

Name	Type	Required	Default	Description
doc_type	string	Yes		文档类型：readme/api/function/module/changelog/tutorial/architecture
path	string	Yes		代码路径（相对工作目录）
language	string	No	auto	代码语言：go/python/javascript/typescript/auto（默认auto）
output_format	string	No	markdown	输出格式：markdown/json（默认markdown）
depth	int	No	3	文档深度1-5（默认3）
auto_update	bool	No	false	是否自动更新现有文档（默认false）

---

doc_parse

Parse documents (PDF/images/HTML) into text, LaTeX, or HTML table format

• Input: string - document text (source=text), base64-encoded image/PDF (source=base64), or URL (source=URL)
• Output: string - parsed document content in the requested output format

Parameters

Name	Type	Required	Default	Description
source	string	No	text	Input source type: text
output_format	string	No	text	Output format: text
extract_tables	bool	No	false	Extract markdown tables and return their count/content (default: false)
extract_formulas	bool	No	false	Extract LaTeX formulas ($...$, $...$) and return their list (default: false)
lang	string	No	auto	Document language hint: zh
api_endpoint	string	No		OCR API endpoint URL (optional). When set with api_key, calls external OCR (e.g. OvisOCR2)
api_key	string	No		OCR API key (optional)

---

drone

Control MAVLink-compatible drones (PX4/ArduPilot) via HTTP bridge. Supports arm/disarm, takeoff, land, RTL, waypoint mission upload, telemetry polling, patrol, survey, orbit, follow-me, and camera capture. Requires a MAVSDK server (drone_bridge.py) running on the drone's companion computer or ground station.

• Input: string - natural language instruction or command description
• Output: string - JSON drone control result with telemetry and status

Parameters

Name	Type	Required	Default	Description
drone_model	string	No	PX4	Drone firmware: PX4, ArduPilot, ArduCopter, ArduPlane, ArduRover, generic (default: PX4)
drone_id	string	No		Unique drone identifier (default: auto-generated)
bridge_host	string	No	127.0.0.1	MAVSDK bridge host IP (default: 127.0.0.1)
bridge_port	string	No	8080	MAVSDK bridge port (default: 50051 for gRPC, 8080 for HTTP)
bridge_token	string	No		Bearer token for bridge authentication (default: none)
action	string	No		Action: arm, disarm, takeoff, land, rtl, hold, goto, mission_start, mission_pause, mission_resume, mission_upload, mission_clear, set_mode, get_telemetry, get_status, get_gps, get_battery, camera, deliver, patrol, survey, orbit, follow
mode	string	No	simulate	Backend mode: simulate (default)
target_altitude_m	float	No	10	Target altitude in meters (default: 10)
target_latitude	float	No		Target latitude for goto/mission waypoints
target_longitude	float	No		Target longitude for goto/mission waypoints
mission_type	string	No	waypoint	Mission type: waypoint, survey, corridor, orbit, patrol (default: waypoint)
waypoints	string	No		JSON array of waypoints [{lat, lon, alt}] for mission_upload
flight_speed_ms	float	No	5	Flight speed in m/s (default: 5)
safety_altitude_m	float	No	3	Minimum safety altitude in meters (default: 3)
max_flight_time_s	string	No	300	Maximum flight time in seconds (default: 300)
geofence_radius_m	float	No	200	Geofence radius in meters (default: 200)
timeout	string	No	15	HTTP request timeout in seconds (default: 15)

---

email_send

Send an email over SMTP (implicit TLS on port 465, STARTTLS elsewhere; plaintext only for loopback relays)

• Input: string - email body (used when the body param is empty)
• Output: string - delivery summary (host, recipients, subject)

Parameters

Name	Type	Required	Default	Description
host	string	Yes		SMTP server hostname or IP (e.g. smtp.gmail.com)
port	int	No	587	SMTP port (465=implicit TLS, 587=STARTTLS; default 587)
from	string	Yes		Sender address (plain email; display names are stripped)
to	string	Yes		Comma-separated recipient addresses (max 50 incl. cc)
cc	string	No		Comma-separated CC addresses (optional)
subject	string	No	aflare notification	Subject line (default 'aflare notification')
body	string	No		Email body (overrides the node input)
username	string	No		SMTP AUTH username (omit for unauthenticated local relays)
password	string	No		SMTP password. Prefer password_env so secrets stay out of workflow files
password_env	string	No		Environment variable holding the SMTP password (e.g. AFLARE_SMTP_PASSWORD)
tls_mode	string	No	auto	TLS strategy: auto (465=implicit TLS, else STARTTLS), starttls (always require STARTTLS), tls (always implicit TLS)
timeout	int	No	30	Dial+dialogue timeout in seconds (default 30, max 120)

---

engineer_skills

预置工程技能包，覆盖前端/后端/DevOps/架构/安全/数据/信创七大领域共 24 项技能。支持技能匹配、应用和版本管理。

• Input: string - task description for skill matching
• Output: string - JSON with skills information

Parameters

Name	Type	Required	Default	Description
action	string	No	list	Action: list/match/apply/get (default: list)
skill_category	string	No		Skill category: frontend/backend/devops/architecture
skill_name	string	No		Skill name
task_description	string	No		Task description for matching (max 5000 chars)
version	string	No	1.0.0	Skill version

---

evaluator

Evaluator agent that scores output against criteria with structured rubrics

• Input: string - the content to evaluate
• Output: string - evaluation scores and justification

Parameters

Name	Type	Required	Default	Description
provider	string	No	ollama	LLM provider (default: ollama)
model	string	No	llama3	Model name (default: llama3)
api_key	string	No		API key
endpoint	string	No		API endpoint URL
rubric	string	No	quality	Evaluation rubric: quality, correctness, completeness, clarity, custom (default: quality)
criteria	string	No		Custom criteria for evaluation (comma-separated dimensions)
scale	string	No	1-10	Score scale: 1-5, 1-10, percentage (default: 1-10)
threshold	string	No		Pass/fail threshold score (e.g. 7)
output_format	string	No	json	Output format: json, markdown, score_only (default: json)

---

execute

Execute shell commands (disabled in safe mode)

• Input: string - stdin for the command
• Output: string - stdout and stderr of the command

Parameters

Name	Type	Required	Default	Description
command	string	Yes		Shell command to execute
dry_run	string	No	false	If true, preview the command without executing (default: false)
timeout	string	No	5m	Command timeout, e.g. 30s, 5m, 1h (default: 5m)

---

fastgpt

Call FastGPT API

• Input: string - user message content
• Output: string - AI response content

Parameters

Name	Type	Required	Default	Description
api_key	string	No		FastGPT API key (or set FASTGPT_API_KEY env var)
app_id	string	No		FastGPT app ID
chat_id	string	No		Chat ID for conversation continuity
endpoint	string	No	https://fastgpt.in/api/v1	API base URL (or set FASTGPT_BASE_URL env var)
system	string	No		System prompt

---

fetch_url

Fetch content from a URL

• Input: string - optional URL (overrides url param)
• Output: string - content of the URL

Parameters

Name	Type	Required	Default	Description
url	string	No		URL to fetch
mode	string	No	text	Extraction mode: text, markdown, html, main_content
timeout	int	No	30	Request timeout in seconds

---

file_read

Read content from a file. Automatically redacts secrets (API keys, tokens, .env files) by default for privacy — set redact=false to disable. With a connector (files/notes) the path is resolved inside the connector's root and its include/max_bytes ceilings apply.

• Input: string - not used
• Output: string - file content (with secrets redacted by default)

Parameters

Name	Type	Required	Default	Description
path	string	Yes		File path to read from. Relative to the working directory, or relative to the connector root when connector is set.
connector	string	No		Named files/notes connector (aflare connector add --type files
redact	string	No	true	Redact secrets in output: true (default) / false. When true, .env and credential files are fully masked; other files have known secret patterns masked.

---

file_watch

Polls a file or directory for create/modify/delete events and returns them as JSON. Suitable for log-monitor and file-organizer workflows.

• Input: string - not used
• Output: string - JSON with watched_path, duration, events_collected, and events array

Parameters

Name	Type	Required	Default	Description
path	string	Yes		要监控的文件或目录路径
duration	string	No	30s	监控持续时间（如 30s, 5m）
interval	string	No	1s	轮询间隔（如 1s, 500ms）
events	string	No	create,modify,delete	关注的事件类型，逗号分隔：create,modify,delete
pattern	string	No	*	文件名 glob 匹配模式
max_events	string	No	1000	最大收集事件数（防 DoS）

---

file_write

Write content to a file

• Input: string - content to write to the file
• Output: string - confirmation message

Parameters

Name	Type	Required	Default	Description
path	string	Yes		File path to write to. Relative to the working directory, or relative to the connector root when connector is set.
content	string	No		Content to write; defaults to the step input (previous step output). Expressions like {{var.x}} are evaluated before the node runs.
connector	string	No		Named files/notes connector. Must be registered with --writable; paths resolve inside its root and its include allowlist applies.
mode	string	No	write	Write mode: write (default) or append

---

files_list

List files under a files/notes connector root (relative paths + sizes). Skips dotfiles/dot-directories and symlinks. The connector's include allowlist applies. Use it to discover paths before file_read.

• Input: string - not used
• Output: string - JSON {files: [{path, bytes}], count, truncated}

Parameters

Name	Type	Required	Default	Description
connector	string	Yes		Named files/notes connector whose root is listed.
pattern	string	No	*/	Glob matched against paths relative to the connector root, e.g. "notes/.md" (single level) or "**/.md" (any depth). Default: all files.
max_entries	string	No	200	Max entries to return (default 200, hard cap 1000).

---

fireworks

Call Fireworks AI LLM API

• Input: string - user message content
• Output: string - AI response content

Parameters

Name	Type	Required	Default	Description
model	string	No	accounts/fireworks/models/llama-v3p3-70b-instruct	Model name (default: accounts/fireworks/models/llama-v3p3-70b-instruct)
api_key	string	No		Fireworks AI API key (or set FIREWORKS_API_KEY env var)
endpoint	string	No	https://api.fireworks.ai/inference/v1	API base URL (default: https://api.fireworks.ai/inference/v1)
system	string	No		System prompt
temperature	string	No		Sampling temperature 0.0-2.0 (default: provider default)
max_tokens	string	No		Max tokens to generate
top_p	string	No		Nucleus sampling probability mass 0.0-1.0
frequency_penalty	string	No		Penalty for repeated tokens -2.0 to 2.0
presence_penalty	string	No		Penalty for new tokens -2.0 to 2.0
stop	string	No		Stop sequences (comma-separated, e.g. '\n,END')
seed	string	No		Random seed for deterministic sampling (int)
response_format	string	No		Structured output: 'json_object' or 'json_schema:<schema_json>'
tools	string	No		JSON array of tool definitions for function calling
tool_choice	string	No		Tool selection: 'none', 'auto', or JSON object
user	string	No		End-user identifier for provider-side abuse monitoring

---

gemini

Call Google Gemini LLM API

• Input: string - user message content
• Output: string - AI response content

Parameters

Name	Type	Required	Default	Description
model	string	No	gemini-2.0-flash	Model name (default: gemini-2.0-flash)
api_key	string	No		Google Gemini API key (or set GEMINI_API_KEY env var)
endpoint	string	No	https://generativelanguage.googleapis.com/v1beta/openai	API base URL (default: https://generativelanguage.googleapis.com/v1beta/openai)
system	string	No		System prompt
temperature	string	No		Sampling temperature 0.0-2.0 (default: provider default)
max_tokens	string	No		Max tokens to generate
top_p	string	No		Nucleus sampling probability mass 0.0-1.0
frequency_penalty	string	No		Penalty for repeated tokens -2.0 to 2.0
presence_penalty	string	No		Penalty for new tokens -2.0 to 2.0
stop	string	No		Stop sequences (comma-separated, e.g. '\n,END')
seed	string	No		Random seed for deterministic sampling (int)
response_format	string	No		Structured output: 'json_object' or 'json_schema:<schema_json>'
tools	string	No		JSON array of tool definitions for function calling
tool_choice	string	No		Tool selection: 'none', 'auto', or JSON object
user	string	No		End-user identifier for provider-side abuse monitoring

---

glm

Call GLM LLM API

• Input: string - user message content
• Output: string - AI response content

Parameters

Name	Type	Required	Default	Description
model	string	No	glm-4	Model name (default: glm-4)
api_key	string	No		GLM API key (or set GLM_API_KEY env var)
endpoint	string	No	https://open.bigmodel.cn/api/paas/v4	API base URL (default: https://open.bigmodel.cn/api/paas/v4)
system	string	No		System prompt
temperature	string	No		Sampling temperature 0.0-2.0 (default: provider default)
max_tokens	string	No		Max tokens to generate
top_p	string	No		Nucleus sampling probability mass 0.0-1.0
frequency_penalty	string	No		Penalty for repeated tokens -2.0 to 2.0
presence_penalty	string	No		Penalty for new tokens -2.0 to 2.0
stop	string	No		Stop sequences (comma-separated, e.g. '\n,END')
seed	string	No		Random seed for deterministic sampling (int)
response_format	string	No		Structured output: 'json_object' or 'json_schema:<schema_json>'
tools	string	No		JSON array of tool definitions for function calling
tool_choice	string	No		Tool selection: 'none', 'auto', or JSON object
user	string	No		End-user identifier for provider-side abuse monitoring

---

groq

Call Groq LLM API

• Input: string - user message content
• Output: string - AI response content

Parameters

Name	Type	Required	Default	Description
model	string	No	llama-3.3-70b-versatile	Model name (default: llama-3.3-70b-versatile)
api_key	string	No		Groq API key (or set GROQ_API_KEY env var)
endpoint	string	No	https://api.groq.com/openai/v1	API base URL (default: https://api.groq.com/openai/v1)
system	string	No		System prompt
temperature	string	No		Sampling temperature 0.0-2.0 (default: provider default)
max_tokens	string	No		Max tokens to generate
top_p	string	No		Nucleus sampling probability mass 0.0-1.0
frequency_penalty	string	No		Penalty for repeated tokens -2.0 to 2.0
presence_penalty	string	No		Penalty for new tokens -2.0 to 2.0
stop	string	No		Stop sequences (comma-separated, e.g. '\n,END')
seed	string	No		Random seed for deterministic sampling (int)
response_format	string	No		Structured output: 'json_object' or 'json_schema:<schema_json>'
tools	string	No		JSON array of tool definitions for function calling
tool_choice	string	No		Tool selection: 'none', 'auto', or JSON object
user	string	No		End-user identifier for provider-side abuse monitoring

---

http_request

Make HTTP requests with custom method, headers, and body

• Input: string - request body (overrides body param)
• Output: string - response body

Parameters

Name	Type	Required	Default	Description
url	string	Yes		Target URL (with connector: path relative to the connector's base URL)
connector	string	No		Named http connector (aflare connector add --type http): pins the base URL and injects auth
method	string	No	GET	HTTP method: GET, POST, PUT, DELETE, PATCH
headers	string	No		JSON-encoded headers
body	string	No		Request body
timeout	int	No	30	Request timeout in seconds
rate_limit_rps	float	No	0	Max requests per second per host (0=unlimited)
rate_limit_burst	int	No		Token-bucket burst size (default=ceil(rate_limit_rps))
rate_limit_key	string	No		Explicit bucket key overriding URL.Host; set when multiple domain aliases resolve to the same backend so they share one bucket (M-9)
max_retries	int	No	0	Max retry attempts on transient failures (default 0=no retry)
retry_backoff_ms	int	No	100	Initial retry backoff in ms (default 100)
retry_max_backoff_ms	int	No	5000	Max retry backoff cap in ms (default 5000)
retry_on_status	string	No	429,500,502,503,504	Comma-separated retryable status codes (default 429,500,502,503,504)

---

human_in_loop

Human approval gate — pauses workflow for human review and approval before continuing

• Input: string - the content/data to present for human review
• Output: string - approved content (or original if approved)

Parameters

Name	Type	Required	Default	Description
mode	string	No	file	Approval mode: file, env, stdin, auto_approve (default: file)
approval_file	string	No	aflare-approval	Path to approval flag file (mode=file). Must not be a dotfile or carry a forbidden extension — the path goes through the standard write-path security validation.
approval_env	string	No	AFLARE_APPROVED	Environment variable to check for approval (mode=env)
prompt	string	No		Custom prompt message for the human reviewer
on_approve	string	No	original	What to output on approve: original, modified, passthrough (default: original)

---

hunyuan

Call Tencent Hunyuan LLM API

• Input: string - user message content
• Output: string - AI response content

Parameters

Name	Type	Required	Default	Description
model	string	No	hunyuan-pro	Model name (default: hunyuan-pro)
api_key	string	No		Tencent Hunyuan API key (or set HUNYUAN_API_KEY env var)
endpoint	string	No	https://api.hunyuan.cloud.tencent.com/v1	API base URL (default: https://api.hunyuan.cloud.tencent.com/v1)
system	string	No		System prompt
temperature	string	No		Sampling temperature 0.0-2.0 (default: provider default)
max_tokens	string	No		Max tokens to generate
top_p	string	No		Nucleus sampling probability mass 0.0-1.0
frequency_penalty	string	No		Penalty for repeated tokens -2.0 to 2.0
presence_penalty	string	No		Penalty for new tokens -2.0 to 2.0
stop	string	No		Stop sequences (comma-separated, e.g. '\n,END')
seed	string	No		Random seed for deterministic sampling (int)
response_format	string	No		Structured output: 'json_object' or 'json_schema:<schema_json>'
tools	string	No		JSON array of tool definitions for function calling
tool_choice	string	No		Tool selection: 'none', 'auto', or JSON object
user	string	No		End-user identifier for provider-side abuse monitoring

---

hygon

Call Hygon DCU LLM API

• Input: string - user message content
• Output: string - AI response content

Parameters

Name	Type	Required	Default	Description
model	string	No	qwen2.5-7b	Model name (default: qwen2.5-7b)
api_key	string	No		Hygon DCU API key (or set HYGON_API_KEY env var)
endpoint	string	No	http://localhost:8082/v1	API base URL (default: http://localhost:8082/v1)
system	string	No		System prompt
temperature	string	No		Sampling temperature 0.0-2.0 (default: provider default)
max_tokens	string	No		Max tokens to generate
top_p	string	No		Nucleus sampling probability mass 0.0-1.0
frequency_penalty	string	No		Penalty for repeated tokens -2.0 to 2.0
presence_penalty	string	No		Penalty for new tokens -2.0 to 2.0
stop	string	No		Stop sequences (comma-separated, e.g. '\n,END')
seed	string	No		Random seed for deterministic sampling (int)
response_format	string	No		Structured output: 'json_object' or 'json_schema:<schema_json>'
tools	string	No		JSON array of tool definitions for function calling
tool_choice	string	No		Tool selection: 'none', 'auto', or JSON object
user	string	No		End-user identifier for provider-side abuse monitoring

---

ima

Call IMA Copilot LLM API

• Input: string - user message content
• Output: string - AI response content

Parameters

Name	Type	Required	Default	Description
model	string	No		Model name (default: )
api_key	string	No		IMA Copilot API key (or set IMA_API_KEY env var)
endpoint	string	No		API base URL (default: )
system	string	No		System prompt
temperature	string	No		Sampling temperature 0.0-2.0 (default: provider default)
max_tokens	string	No		Max tokens to generate
top_p	string	No		Nucleus sampling probability mass 0.0-1.0
frequency_penalty	string	No		Penalty for repeated tokens -2.0 to 2.0
presence_penalty	string	No		Penalty for new tokens -2.0 to 2.0
stop	string	No		Stop sequences (comma-separated, e.g. '\n,END')
seed	string	No		Random seed for deterministic sampling (int)
response_format	string	No		Structured output: 'json_object' or 'json_schema:<schema_json>'
tools	string	No		JSON array of tool definitions for function calling
tool_choice	string	No		Tool selection: 'none', 'auto', or JSON object
user	string	No		End-user identifier for provider-side abuse monitoring

---

internlm

Call InternLM LLM API

• Input: string - user message content
• Output: string - AI response content

Parameters

Name	Type	Required	Default	Description
model	string	No	internlm3-latest	Model name (default: internlm3-latest)
api_key	string	No		InternLM API key (or set INTERNLM_API_KEY env var)
endpoint	string	No	https://internlm-chat.intern-ai.org.cn/api/v1	API base URL (default: https://internlm-chat.intern-ai.org.cn/api/v1)
system	string	No		System prompt
temperature	string	No		Sampling temperature 0.0-2.0 (default: provider default)
max_tokens	string	No		Max tokens to generate
top_p	string	No		Nucleus sampling probability mass 0.0-1.0
frequency_penalty	string	No		Penalty for repeated tokens -2.0 to 2.0
presence_penalty	string	No		Penalty for new tokens -2.0 to 2.0
stop	string	No		Stop sequences (comma-separated, e.g. '\n,END')
seed	string	No		Random seed for deterministic sampling (int)
response_format	string	No		Structured output: 'json_object' or 'json_schema:<schema_json>'
tools	string	No		JSON array of tool definitions for function calling
tool_choice	string	No		Tool selection: 'none', 'auto', or JSON object
user	string	No		End-user identifier for provider-side abuse monitoring

---

json_parse

Parse and extract JSON data

• Input: string - JSON string to parse (a leading "HTTP \n" status line, as emitted by http_request, is tolerated and stripped)
• Output: string - extracted value or pretty-printed JSON

Parameters

Name	Type	Required	Default	Description
path	string	No		Dot-notation path to extract (e.g. data.items.[0].name). If omitted, pretty-prints entire JSON.

---

kimi

Call Kimi LLM API

• Input: string - user message content
• Output: string - AI response content

Parameters

Name	Type	Required	Default	Description
model	string	No	kimi-k3	Model name (default: kimi-k3)
api_key	string	No		Kimi API key (or set KIMI_API_KEY env var)
endpoint	string	No	https://api.moonshot.cn/v1	API base URL (default: https://api.moonshot.cn/v1)
system	string	No		System prompt
temperature	string	No		Sampling temperature 0.0-2.0 (default: provider default)
max_tokens	string	No		Max tokens to generate
top_p	string	No		Nucleus sampling probability mass 0.0-1.0
frequency_penalty	string	No		Penalty for repeated tokens -2.0 to 2.0
presence_penalty	string	No		Penalty for new tokens -2.0 to 2.0
stop	string	No		Stop sequences (comma-separated, e.g. '\n,END')
seed	string	No		Random seed for deterministic sampling (int)
response_format	string	No		Structured output: 'json_object' or 'json_schema:<schema_json>'
tools	string	No		JSON array of tool definitions for function calling
tool_choice	string	No		Tool selection: 'none', 'auto', or JSON object
user	string	No		End-user identifier for provider-side abuse monitoring

---

knowledge_graph

Knowledge graph node - extract entities/relations, build graph, query and traverse

• Input: string - text to extract knowledge from, or a query for an existing graph
• Output: string - knowledge graph data or query results

Parameters

Name	Type	Required	Default	Description
action	string	No	extract	Action: extract, extract_llm, query, traverse, stats, visualize (default: extract). extract_llm calls an LLM for higher-quality entity/relation extraction.
graph_path	string	No		Path to save/load the graph JSON file
query	string	No		Query for search/traverse (entity name or relation type)
max_depth	string	No	2	Max traversal depth (default: 2)
top_k	string	No	10	Max results to return (default: 10)
format	string	No	markdown	Output format: json, markdown, mermaid (default: markdown)
provider	string	No	openai	LLM provider for extract_llm (default: openai)
model	string	No		LLM model name for extract_llm
api_key	string	No		LLM API key for extract_llm (or set _API_KEY env var)
endpoint	string	No		LLM API base URL for extract_llm
language	string	No	en	Prompt language hint for extract_llm: en or zh (default: en)
session_id	string	No		C-3: when set with memory_key, links extracted entities to that memory entry
memory_key	string	No		C-3: memory entry key to link extracted entities to

---

llm_router

Smart LLM router that automatically selects the best provider with fallback, quota tracking, and cost optimization

• Input: string - user message content to send to LLM
• Output: string - AI response from the selected provider

Parameters

Name	Type	Required	Default	Description
system	string	No		System prompt for the LLM
strategy	string	No	priority	Routing strategy: priority, cost, latency, pareto, round_robin, random (default: priority)
max_retries	string	No	3	Maximum number of fallback attempts (default: 3)
show_provider	string	No	false	Show which provider was used in output (default: false)
show_stats	string	No	false	Show router statistics in output (default: false)

---

memory

AI Agent memory infrastructure with session-isolated persistent knowledge graph engine. Supports multi-session parallel memory, short/medium/long term memory, cross-session long-term memory, and memory usage monitoring.

• Input: string - memory content to store or query for retrieval
• Output: string - JSON with memory operations result, entries, or statistics

Parameters

Name	Type	Required	Default	Description
operation	string	No	store	Operation: store/retrieve/delete/search/harness_search/summary/forget/transfer/merge/inkling_retrieve/list_sessions/session_stats/global_stats/link_kg/expand_kg/compress (default: store)
session_id	string	No	default	Session ID for isolated memory (default: default)
key	string	No		Memory key for storage/retrieval/link_kg
value	string	No		Memory value/content
level	string	No	medium	Memory level: short/medium/long (default: medium)
type	string	No	fact	Memory type: fact/concept/experience/preference/relationship/task/context (default: fact)
tags	string	No		Comma-separated tags for categorization
ttl_hours	int	No	72	Time to live in hours (default: 72)
confidence	float	No	0.8	Confidence level 0.0-1.0 (default: 0.8)
query	string	No		Search query for retrieval/search/harness_search/expand_kg operations
top_k	int	No	10	Number of results to return (1-100, default: 10)
threshold	float	No	0.5	Similarity threshold 0.0-1.0 (default: 0.5)
source	string	No		Source identifier for the memory
kg_entities	string	No		link_kg: comma-separated KG entity names to link to the memory key
token_budget	int	No	2000	compress: max tokens to retain after compression (default: 2000)
min_confidence	float	No	0.5	compress: entries below this confidence are candidates for compression (default: 0.5)

---

mimo

Call MiMo LLM API

• Input: string - user message content
• Output: string - AI response content

Parameters

Name	Type	Required	Default	Description
model	string	No	mimo-v2.5-pro	Model name (default: mimo-v2.5-pro)
api_key	string	No		MiMo API key (or set MIMO_API_KEY env var)
endpoint	string	No	https://api.xiaomimimo.com/v1	API base URL (default: https://api.xiaomimimo.com/v1)
system	string	No		System prompt
temperature	string	No		Sampling temperature 0.0-2.0 (default: provider default)
max_tokens	string	No		Max tokens to generate
top_p	string	No		Nucleus sampling probability mass 0.0-1.0
frequency_penalty	string	No		Penalty for repeated tokens -2.0 to 2.0
presence_penalty	string	No		Penalty for new tokens -2.0 to 2.0
stop	string	No		Stop sequences (comma-separated, e.g. '\n,END')
seed	string	No		Random seed for deterministic sampling (int)
response_format	string	No		Structured output: 'json_object' or 'json_schema:<schema_json>'
tools	string	No		JSON array of tool definitions for function calling
tool_choice	string	No		Tool selection: 'none', 'auto', or JSON object
user	string	No		End-user identifier for provider-side abuse monitoring

---

minimax

Call MiniMax LLM API

• Input: string - user message content
• Output: string - AI response content

Parameters

Name	Type	Required	Default	Description
model	string	No	abab6.5s-chat	Model name (default: abab6.5s-chat)
api_key	string	No		MiniMax API key (or set MINIMAX_API_KEY env var)
endpoint	string	No	https://api.minimax.chat/v1	API base URL (default: https://api.minimax.chat/v1)
system	string	No		System prompt
temperature	string	No		Sampling temperature 0.0-2.0 (default: provider default)
max_tokens	string	No		Max tokens to generate
top_p	string	No		Nucleus sampling probability mass 0.0-1.0
frequency_penalty	string	No		Penalty for repeated tokens -2.0 to 2.0
presence_penalty	string	No		Penalty for new tokens -2.0 to 2.0
stop	string	No		Stop sequences (comma-separated, e.g. '\n,END')
seed	string	No		Random seed for deterministic sampling (int)
response_format	string	No		Structured output: 'json_object' or 'json_schema:<schema_json>'
tools	string	No		JSON array of tool definitions for function calling
tool_choice	string	No		Tool selection: 'none', 'auto', or JSON object
user	string	No		End-user identifier for provider-side abuse monitoring

---

mistral

Call Mistral LLM API

• Input: string - user message content
• Output: string - AI response content

Parameters

Name	Type	Required	Default	Description
model	string	No	mistral-large-latest	Model name (default: mistral-large-latest)
api_key	string	No		Mistral API key (or set MISTRAL_API_KEY env var)
endpoint	string	No	https://api.mistral.ai/v1	API base URL (default: https://api.mistral.ai/v1)
system	string	No		System prompt
temperature	string	No		Sampling temperature 0.0-2.0 (default: provider default)
max_tokens	string	No		Max tokens to generate
top_p	string	No		Nucleus sampling probability mass 0.0-1.0
frequency_penalty	string	No		Penalty for repeated tokens -2.0 to 2.0
presence_penalty	string	No		Penalty for new tokens -2.0 to 2.0
stop	string	No		Stop sequences (comma-separated, e.g. '\n,END')
seed	string	No		Random seed for deterministic sampling (int)
response_format	string	No		Structured output: 'json_object' or 'json_schema:<schema_json>'
tools	string	No		JSON array of tool definitions for function calling
tool_choice	string	No		Tool selection: 'none', 'auto', or JSON object
user	string	No		End-user identifier for provider-side abuse monitoring

---

multimodal

Multimodal node for image analysis, OCR, and audio transcription using vision-capable LLMs

• Input: string - the question or instruction about the media
• Output: string - analysis result from the multimodal model

Parameters

Name	Type	Required	Default	Description
mode	string	No	describe	Mode: image, ocr, describe, compare (default: describe)
image_path	string	No		Path to image file (local path or URL)
image_paths	string	No		Comma-separated paths for compare mode
lang	string	No	eng+chi_sim	OCR languages for tesseract (default: eng+chi_sim)
provider	string	No	openai	LLM provider with vision support (default: openai)
model	string	No	gpt-4o	Vision model name (default: gpt-4o)
api_key	string	No		API key
endpoint	string	No		API endpoint
detail	string	No	auto	Image detail level: low, high, auto (default: auto)
output_format	string	No	markdown	Output format: text, json, markdown (default: markdown)

---

notify

Send notifications (stdout, stderr, slack, discord, telegram, feishu, dingtalk, wecom, webhook)

• Input: string - message to notify (used if message param is empty)
• Output: string - the notification message

Parameters

Name	Type	Required	Default	Description
channel	string	No	stdout	Notification channel: stdout, stderr, slack, discord, telegram, feishu, dingtalk, wecom, webhook (default: stdout)
message	string	No		Notification message (overrides input)
url	string	No		Webhook URL for slack/discord/webhook/feishu/dingtalk/wecom, or Telegram API base (required for external channels)
webhook_url	string	No		Deprecated: use url instead
token	string	No		Bot token (required when channel=telegram)
chat_id	string	No		Telegram chat ID (required when channel=telegram)
username	string	No		Discord webhook username (optional)
method	string	No	POST	HTTP method for webhook: GET/POST/PUT (default: POST)
headers	string	No		JSON headers for webhook (optional)
body	string	No		Custom body for webhook (optional)

---

nvidia

Call NVIDIA NIM LLM API

• Input: string - user message content
• Output: string - AI response content

Parameters

Name	Type	Required	Default	Description
model	string	No	meta/llama-3.3-70b-instruct	Model name (default: meta/llama-3.3-70b-instruct)
api_key	string	No		NVIDIA NIM API key (or set NVIDIA_API_KEY env var)
endpoint	string	No	https://integrate.api.nvidia.com/v1	API base URL (default: https://integrate.api.nvidia.com/v1)
system	string	No		System prompt
temperature	string	No		Sampling temperature 0.0-2.0 (default: provider default)
max_tokens	string	No		Max tokens to generate
top_p	string	No		Nucleus sampling probability mass 0.0-1.0
frequency_penalty	string	No		Penalty for repeated tokens -2.0 to 2.0
presence_penalty	string	No		Penalty for new tokens -2.0 to 2.0
stop	string	No		Stop sequences (comma-separated, e.g. '\n,END')
seed	string	No		Random seed for deterministic sampling (int)
response_format	string	No		Structured output: 'json_object' or 'json_schema:<schema_json>'
tools	string	No		JSON array of tool definitions for function calling
tool_choice	string	No		Tool selection: 'none', 'auto', or JSON object
user	string	No		End-user identifier for provider-side abuse monitoring

---

office

Read .docx/.xlsx/.pptx documents (text, tables, slides) using pure-Go OOXML parsing

• Input: string - path to the office document (.docx/.xlsx/.pptx)
• Output: string - extracted content in the requested output format

Parameters

Name	Type	Required	Default	Description
path	string	No		Path to the office document (overrides input if set)
format	string	No		Source format hint: docx
output	string	No	markdown	Output format: text
sheet	string	No		xlsx only: sheet name to read (default: all sheets)
max_rows	int	No	1000	xlsx only: max rows per sheet (default: 1000, 0 = unlimited)

---

ollama

Call Ollama local LLM server

• Input: string - user prompt content (used when prompt param is not provided)
• Output: string - model response content

Parameters

Name	Type	Required	Default	Description
model	string	No	llama3	Model name (default: llama3)
endpoint	string	No	http://localhost:11434	Ollama server URL (default: http://localhost:11434)
prompt	string	No		Prompt to send to Ollama (if not provided, uses input)

---

openai

Call OpenAI LLM API

• Input: string - user message content
• Output: string - AI response content

Parameters

Name	Type	Required	Default	Description
model	string	No	gpt-3.5-turbo	Model name (default: gpt-3.5-turbo)
api_key	string	No		OpenAI API key (or set OPENAI_API_KEY env var)
endpoint	string	No	https://api.openai.com/v1	API base URL (default: https://api.openai.com/v1)
system	string	No		System prompt
temperature	string	No		Sampling temperature 0.0-2.0 (default: provider default)
max_tokens	string	No		Max tokens to generate
top_p	string	No		Nucleus sampling probability mass 0.0-1.0
frequency_penalty	string	No		Penalty for repeated tokens -2.0 to 2.0
presence_penalty	string	No		Penalty for new tokens -2.0 to 2.0
stop	string	No		Stop sequences (comma-separated, e.g. '\n,END')
seed	string	No		Random seed for deterministic sampling (int)
response_format	string	No		Structured output: 'json_object' or 'json_schema:<schema_json>'
tools	string	No		JSON array of tool definitions for function calling
tool_choice	string	No		Tool selection: 'none', 'auto', or JSON object
user	string	No		End-user identifier for provider-side abuse monitoring

---

openrouter

Call OpenRouter LLM API

• Input: string - user message content
• Output: string - AI response content

Parameters

Name	Type	Required	Default	Description
model	string	No	openrouter/auto	Model name (default: openrouter/auto)
api_key	string	No		OpenRouter API key (or set OPENROUTER_API_KEY env var)
endpoint	string	No	https://openrouter.ai/api/v1	API base URL (default: https://openrouter.ai/api/v1)
system	string	No		System prompt
temperature	string	No		Sampling temperature 0.0-2.0 (default: provider default)
max_tokens	string	No		Max tokens to generate
top_p	string	No		Nucleus sampling probability mass 0.0-1.0
frequency_penalty	string	No		Penalty for repeated tokens -2.0 to 2.0
presence_penalty	string	No		Penalty for new tokens -2.0 to 2.0
stop	string	No		Stop sequences (comma-separated, e.g. '\n,END')
seed	string	No		Random seed for deterministic sampling (int)
response_format	string	No		Structured output: 'json_object' or 'json_schema:<schema_json>'
tools	string	No		JSON array of tool definitions for function calling
tool_choice	string	No		Tool selection: 'none', 'auto', or JSON object
user	string	No		End-user identifier for provider-side abuse monitoring

---

perplexity

Call Perplexity LLM API

• Input: string - user message content
• Output: string - AI response content

Parameters

Name	Type	Required	Default	Description
model	string	No	sonar	Model name (default: sonar)
api_key	string	No		Perplexity API key (or set PERPLEXITY_API_KEY env var)
endpoint	string	No	https://api.perplexity.ai	API base URL (default: https://api.perplexity.ai)
system	string	No		System prompt
temperature	string	No		Sampling temperature 0.0-2.0 (default: provider default)
max_tokens	string	No		Max tokens to generate
top_p	string	No		Nucleus sampling probability mass 0.0-1.0
frequency_penalty	string	No		Penalty for repeated tokens -2.0 to 2.0
presence_penalty	string	No		Penalty for new tokens -2.0 to 2.0
stop	string	No		Stop sequences (comma-separated, e.g. '\n,END')
seed	string	No		Random seed for deterministic sampling (int)
response_format	string	No		Structured output: 'json_object' or 'json_schema:<schema_json>'
tools	string	No		JSON array of tool definitions for function calling
tool_choice	string	No		Tool selection: 'none', 'auto', or JSON object
user	string	No		End-user identifier for provider-side abuse monitoring

---

pipeline

Dependency-based parallel workflow executor: steps run as soon as their dependencies are met, no global barriers (Tunix-inspired async rollout)

• Input: string - YAML or JSON pipeline configuration with steps and dependencies. Step fields: name, node, input, params, depends_on (list), input_from (list). If a step fails, all transitive downstream steps are cascade-skipped (results carry skipped: true); independent branches still run
• Output: string - JSON with execution results, timings, and errors; skipped steps appear with skipped: true and error "skipped: upstream step ... failed"

Parameters

Name	Type	Required	Default	Description
timeout	string	No	300	Timeout in seconds (default: 300)
format	string	No	auto	Input format: json

---

planner

Task decomposition agent that breaks complex goals into actionable steps

• Input: string - the complex task or goal to plan for
• Output: string - JSON array of planned steps with descriptions and dependencies

Parameters

Name	Type	Required	Default	Description
provider	string	No	ollama	LLM provider (default: ollama)
model	string	No	llama3	Model name (default: llama3)
api_key	string	No		API key
endpoint	string	No		API endpoint URL
max_steps	string	No	10	Maximum number of steps to plan (default: 10)
context	string	No		Additional context or constraints for the plan
auto_clarify	string	No	false	Run ACQUIRE-style clarification before planning (default: false). If task is ambiguous, ask clarifying questions first.
clarify_threshold	string	No	70	Confidence threshold for auto-clarification 0-100 (default: 70)
clarify_max_questions	string	No	3	Max clarification questions (default: 3)

---

preference

User preference memory: store, retrieve, and learn user habits across sessions (MemSlides-inspired user profiling)

• Input: string - input depends on operation (value to set, key to get, etc.)
• Output: string - result of the operation

Parameters

Name	Type	Required	Default	Description
operation	string	No	get	get
user_id	string	No	default	User identifier (default: default)
category	string	No	custom	coding_style
key	string	No		Preference key name
value	string	No		Preference value (for set/learn operations)
confidence	string	No		Confidence 0-1, default 0.6 for learn, 1.0 for set
source	string	No	explicit	Where this preference came from (explicit

---

qianfan

Call Baidu Qianfan LLM API

• Input: string - user message content
• Output: string - AI response content

Parameters

Name	Type	Required	Default	Description
model	string	No	ernie-4.5-turbo-128k	Model name (default: ernie-4.5-turbo-128k)
api_key	string	No		Baidu Qianfan API key (or set QIANFAN_API_KEY env var)
endpoint	string	No	https://qianfan.baidubce.com/v2	API base URL (default: https://qianfan.baidubce.com/v2)
system	string	No		System prompt
temperature	string	No		Sampling temperature 0.0-2.0 (default: provider default)
max_tokens	string	No		Max tokens to generate
top_p	string	No		Nucleus sampling probability mass 0.0-1.0
frequency_penalty	string	No		Penalty for repeated tokens -2.0 to 2.0
presence_penalty	string	No		Penalty for new tokens -2.0 to 2.0
stop	string	No		Stop sequences (comma-separated, e.g. '\n,END')
seed	string	No		Random seed for deterministic sampling (int)
response_format	string	No		Structured output: 'json_object' or 'json_schema:<schema_json>'
tools	string	No		JSON array of tool definitions for function calling
tool_choice	string	No		Tool selection: 'none', 'auto', or JSON object
user	string	No		End-user identifier for provider-side abuse monitoring

---

qwen

Call Qwen LLM API

• Input: string - user message content
• Output: string - AI response content

Parameters

Name	Type	Required	Default	Description
model	string	No	qwen-turbo	Model name (default: qwen-turbo)
api_key	string	No		Qwen API key (or set QWEN_API_KEY env var)
endpoint	string	No	https://dashscope.aliyuncs.com/compatible-mode/v1	API base URL (default: https://dashscope.aliyuncs.com/compatible-mode/v1)
system	string	No		System prompt
temperature	string	No		Sampling temperature 0.0-2.0 (default: provider default)
max_tokens	string	No		Max tokens to generate
top_p	string	No		Nucleus sampling probability mass 0.0-1.0
frequency_penalty	string	No		Penalty for repeated tokens -2.0 to 2.0
presence_penalty	string	No		Penalty for new tokens -2.0 to 2.0
stop	string	No		Stop sequences (comma-separated, e.g. '\n,END')
seed	string	No		Random seed for deterministic sampling (int)
response_format	string	No		Structured output: 'json_object' or 'json_schema:<schema_json>'
tools	string	No		JSON array of tool definitions for function calling
tool_choice	string	No		Tool selection: 'none', 'auto', or JSON object
user	string	No		End-user identifier for provider-side abuse monitoring

---

rag

Retrieval Augmented Generation node - chunk documents, search by query, and assemble context

• Input: string - the query to search for
• Output: string - assembled context from relevant document chunks

Parameters

Name	Type	Required	Default	Description
source	string	Yes		Source: file path, directory path, or text content
source_type	string	No	auto	Type of source: file, dir, text (default: auto)
chunk_size	string	No	1000	Chunk size in characters (default: 1000)
chunk_overlap	string	No	200	Chunk overlap in characters (default: 200)
top_k	string	No	5	Number of top chunks to retrieve (default: 5)
search_method	string	No	keyword	Search method: keyword, hybrid (default: keyword)
include_metadata	string	No	true	Include chunk metadata in output (default: true)

---

reflector

Self-reflection agent that critiques output and iteratively improves it (Reflexion pattern)

• Input: string - the initial output to reflect on and improve
• Output: string - the improved final output

Parameters

Name	Type	Required	Default	Description
provider	string	No	ollama	LLM provider (default: ollama)
model	string	No	llama3	Model name (default: llama3)
api_key	string	No		API key
endpoint	string	No		API endpoint URL
iterations	string	No	2	Number of reflection iterations (1-5, default: 2)
goal	string	No		The original goal/task the output was trying to achieve
reflection_focus	string	No	all	What to reflect on: accuracy, completeness, quality, all (default: all)

---

researcher

Research agent that fetches information from URLs and summarizes findings

• Input: string - the research topic or question
• Output: string - structured research summary with sources

Parameters

Name	Type	Required	Default	Description
provider	string	No	ollama	LLM provider (default: ollama)
model	string	No	llama3	Model name (default: llama3)
api_key	string	No		API key
endpoint	string	No		API endpoint URL
urls	string	No		Comma-separated URLs to research (if not provided, agent will use input topic)
depth	string	No	basic	Research depth: basic, detailed, comprehensive (default: basic)
output_format	string	No	markdown	Output format: markdown, json, summary (default: markdown)

---

search_aggregate

Multi-platform search aggregator with real-signal ranking: Reddit/Twitter/YouTube/HN/GitHub, sorted by votes/comments/shares instead of editorial SEO (last30days-skill inspired)

• Input: string - search query
• Output: string - JSON or formatted ranked results with signal data

Parameters

Name	Type	Required	Default	Description
sources	string	No	reddit,hn,github,news	Comma-separated sources: reddit,twitter,youtube,hn,github,google,weibo,zhihu,bilibili,linkedin,news,finance,academic,shopping,geopolitical,infrastructure,globalevents,energy,supplychain (default: reddit,hn,github,news)
region	string	No	global	Region filter: global,us,eu,asia,cn,mena (default: global)
category	string	No	all	Category filter: politics,economy,technology,military,energy,health,all (default: all)
limit	string	No	10	Max results per source (default: 10)
time_range	string	No	week	Time range: day
sort_by	string	No	signal	signal
min_score	string	No	0	Minimum combined signal score filter (default: 0)
output	string	No	markdown	json

---

session_manager

Multi-session memory management. Create isolated sessions, fork a session from a parent, merge sessions, and share facts across sessions via the shared namespace.

• Input: string - value to store (for shared_put action)
• Output: string - JSON result of the action

Parameters

Name	Type	Required	Default	Description
action	string	Yes	list	create
session_id	string	No		Target session id (required for most actions)
parent	string	No		Parent session id to inherit memory from (create action only)
src	string	No		Source session id (merge action only)
dst	string	No		Destination session id (merge action only)
key	string	No		Memory key (shared_get/shared_put/recall actions)
value	string	No		Memory value (shared_put action). Overrides input when set.
level	string	No	short	Memory level: short
type	string	No	fact	Memory type tag (default fact)
query	string	No		Search query (search action)
top_k	string	No	10	Max search results (default 10)

---

siliconflow

Call SiliconFlow LLM API

• Input: string - user message content
• Output: string - AI response content

Parameters

Name	Type	Required	Default	Description
model	string	No	Qwen/Qwen3-32B	Model name (default: Qwen/Qwen3-32B)
api_key	string	No		SiliconFlow API key (or set SILICONFLOW_API_KEY env var)
endpoint	string	No	https://api.siliconflow.cn/v1	API base URL (default: https://api.siliconflow.cn/v1)
system	string	No		System prompt
temperature	string	No		Sampling temperature 0.0-2.0 (default: provider default)
max_tokens	string	No		Max tokens to generate
top_p	string	No		Nucleus sampling probability mass 0.0-1.0
frequency_penalty	string	No		Penalty for repeated tokens -2.0 to 2.0
presence_penalty	string	No		Penalty for new tokens -2.0 to 2.0
stop	string	No		Stop sequences (comma-separated, e.g. '\n,END')
seed	string	No		Random seed for deterministic sampling (int)
response_format	string	No		Structured output: 'json_object' or 'json_schema:<schema_json>'
tools	string	No		JSON array of tool definitions for function calling
tool_choice	string	No		Tool selection: 'none', 'auto', or JSON object
user	string	No		End-user identifier for provider-side abuse monitoring

---

skill_distill

Distill methodologies from books, videos, podcasts, and documents into callable skills. Supports workflow, decision, analysis, creative, prompt, and checklist skill types.

• Input: string - source content to distill
• Output: string - JSON with distilled skill structure

Parameters

Name	Type	Required	Default	Description
source_type	string	No	article	Source type: book/video/podcast/article/documentation/conversation (default: article)
distill_type	string	No	workflow	Distill type: workflow/decision/analysis/creative/prompt/checklist (default: workflow)
content	string	No		Source content text (max 100000 chars)
skill_name	string	No		Target skill name
max_steps	int	No	10	Max number of steps (default: 10)
quality	string	No	standard	Quality level: basic/standard/expert (default: standard)

---

sql_query

Execute SQL via database/sql. The driver must be registered by the host program. Uses parameterized queries (? or $1) to prevent SQL injection. Read-only by default (SELECT/SHOW/EXPLAIN/PRAGMA only); set read_only=false to allow DML/DDL. Supports a 'schema' action that lists tables and columns. Prefer the connector param (named connection from aflare connector add) over inline driver/dsn — connectors keep credentials out of workflow files and enforce their own read_only/max_rows/timeout ceilings.

• Input: string - SQL query (when action=query and no sql param, input is used as the query)
• Output: string - JSON array of rows (query), or schema description (schema action)

Parameters

Name	Type	Required	Default	Description
connector	string	No		Named connector registered via aflare connector add. When set, driver/dsn/credentials are resolved from the connector and inline driver/dsn must be omitted. Node-level read_only/max_rows/timeout can only tighten the connector's policy.
action	string	No	query	query (default)
driver	string	No		database/sql driver name (e.g. sqlite3, postgres, mysql). Required unless connector is set.
dsn	string	No		Data source name (driver-specific). For SQLite: path to .db file. Required unless connector is set.
sql	string	No		SQL statement. Use ? (mysql/sqlite) or $1,$2 (postgres) placeholders for args.
args	string	No		JSON array of bind parameters, e.g. ["foo", 42]. Optional.
read_only	string	No	true	Reject writes if true (default). Set false to allow INSERT/UPDATE/DELETE/DDL. A read-only connector stays read-only regardless.
max_rows	string	No	1000	Max rows to return (default 1000; a connector's max_rows is the ceiling). Protects against huge result sets.
timeout	string	No	30	Query timeout in seconds (default 30; a connector's timeout is the ceiling).

---

stepfun

Call StepFun LLM API

• Input: string - user message content
• Output: string - AI response content

Parameters

Name	Type	Required	Default	Description
model	string	No	step-2-16k	Model name (default: step-2-16k)
api_key	string	No		StepFun API key (or set STEPFUN_API_KEY env var)
endpoint	string	No	https://api.stepfun.com/v1	API base URL (default: https://api.stepfun.com/v1)
system	string	No		System prompt
temperature	string	No		Sampling temperature 0.0-2.0 (default: provider default)
max_tokens	string	No		Max tokens to generate
top_p	string	No		Nucleus sampling probability mass 0.0-1.0
frequency_penalty	string	No		Penalty for repeated tokens -2.0 to 2.0
presence_penalty	string	No		Penalty for new tokens -2.0 to 2.0
stop	string	No		Stop sequences (comma-separated, e.g. '\n,END')
seed	string	No		Random seed for deterministic sampling (int)
response_format	string	No		Structured output: 'json_object' or 'json_schema:<schema_json>'
tools	string	No		JSON array of tool definitions for function calling
tool_choice	string	No		Tool selection: 'none', 'auto', or JSON object
user	string	No		End-user identifier for provider-side abuse monitoring

---

structured_output

LLM-driven structured output with local JSON Schema validation and self-correction retries

• Input: string - user instruction describing what to produce
• Output: string - JSON string validated against schema

Parameters

Name	Type	Required	Default	Description
schema	string	Yes		JSON Schema (draft-07 subset) the output must conform to. Must be a JSON object with "type":"object" at the root.
schema_name	string	No	output	Optional schema name shown to the model in the prompt (default: "output")
provider	string	No	openai	Provider name for default model/endpoint resolution (default: openai)
model	string	No		Model name (default: provider default)
api_key	string	No		LLM API key (or set _API_KEY env var)
endpoint	string	No		LLM API base URL (default: provider default endpoint)
system	string	No		Additional system prompt prepended to the JSON instruction
temperature	string	No	0	Sampling temperature 0.0-2.0 (default: 0.0 for deterministic output)
max_tokens	string	No		Max tokens to generate
max_retries	string	No	2	Max self-correction retries on parse/validation failure (default: 2)
format_output	string	No	false	If "true", pretty-print the JSON output (default: false, compact)

---

supervisor

Advanced supervisor with MoE routing, MindSearch deep research, 232+ domain specialists, and collaboration templates. External-agent delegations are checkpointed at (agent, subtask) granularity when the run uses --resume: a crashed supervisor step restores its finished delegations and re-dispatches only the unfinished ones (reported as resumed in the results envelope); without checkpointing the behavior is exactly one-shot

• Input: string - the overall goal or task to supervise
• Output: string - structured task plan with delegation and synthesis

Parameters

Name	Type	Required	Default	Description
provider	string	No	ollama	LLM provider (default: ollama)
model	string	No	llama3	Model name (default: llama3)
api_key	string	No		API key for cloud providers
endpoint	string	No		API endpoint URL
specialists	string	No	planner,researcher,critic,evaluator	Comma-separated specialists. Persona roles: planner,researcher,critic,code_review,evaluator,reflector,legal_expert,medical_expert,educational_expert,financial_expert,creative_writer,data_analyst. Registered external agents: prefix with @ (e.g. @codex,@claude,@my-a2a-agent) for real delegation
max_parallel	string	No	4	Max concurrent external-agent delegations; excess subtasks queue (default: 4, max: 16)
fail_on	string	No	none	Overall failure policy for delegated agents: none = never fail the node, failures stay isolated in the results (default); all = fail the node only if every delegation failed; any = fail the node if any delegation failed
delegation_timeout	string	No		Per-delegation timeout for external agents, e.g. 30s, 10m (default: 10m, max: 60m). Agents failing repeatedly trip a per-agent circuit breaker: delegations fail fast with circuit-open instead of burning the full timeout, and recovery is probed via the agent card
strategy	string	No	sequential	Strategy: sequential, parallel, hierarchical, mindsearch, moe, agency, swarm (default: sequential)
output_format	string	No	json	Output format: json, markdown, summary (default: json)
domain	string	No	general	Domain specialization: general,legal,medical,education,finance,creative,tech,business (default: general)
enable_moe	string	No	false	Enable Mixture-of-Experts routing (default: false)
max_depth	string	No	3	Max decomposition depth for hierarchical/mindsearch (default: 3)
subagent_prompts	string	No	true	Inject per-specialist subagent prompt templates into the supervisor context (default: true). Borrows Grok Build's main/subagent prompt hierarchy.
collaboration_template	string	No		Collaboration template: software_development, product_design, data_science, marketing, research, legal_compliance, healthcare, education, finance, game_development, video_production, security_operations, cloud_infrastructure, content_creation, community_management, startup_acceleration, ai_development, design_system, event_management, translation_localization
template_role	string	No	team	Template role to use: team, workflow, review_cycle (default: team)

---

template_render

Render Go templates with input data

• Input: string - input data available as .input in template
• Output: string - rendered template output

Parameters

Name	Type	Required	Default	Description
template	string	No		Inline template string
template_file	string	No		Path to template file (takes precedence over template)

---

together

Call Together AI LLM API

• Input: string - user message content
• Output: string - AI response content

Parameters

Name	Type	Required	Default	Description
model	string	No	meta-llama/Llama-3.3-70B-Instruct-Turbo	Model name (default: meta-llama/Llama-3.3-70B-Instruct-Turbo)
api_key	string	No		Together AI API key (or set TOGETHER_API_KEY env var)
endpoint	string	No	https://api.together.xyz/v1	API base URL (default: https://api.together.xyz/v1)
system	string	No		System prompt
temperature	string	No		Sampling temperature 0.0-2.0 (default: provider default)
max_tokens	string	No		Max tokens to generate
top_p	string	No		Nucleus sampling probability mass 0.0-1.0
frequency_penalty	string	No		Penalty for repeated tokens -2.0 to 2.0
presence_penalty	string	No		Penalty for new tokens -2.0 to 2.0
stop	string	No		Stop sequences (comma-separated, e.g. '\n,END')
seed	string	No		Random seed for deterministic sampling (int)
response_format	string	No		Structured output: 'json_object' or 'json_schema:<schema_json>'
tools	string	No		JSON array of tool definitions for function calling
tool_choice	string	No		Tool selection: 'none', 'auto', or JSON object
user	string	No		End-user identifier for provider-side abuse monitoring

---

transform

Transform text using string operations

• Input: string - text to transform
• Output: string - transformed text

Parameters

Name	Type	Required	Default	Description
operation	string	No		Transformation operation

---

verify

Agent-as-a-Judge verifier that validates outputs, claims, and results against specified criteria

• Input: string - the content to verify (used as claim if claim param is empty)
• Output: string - verification result with pass/fail, score, or detailed analysis

Parameters

Name	Type	Required	Default	Description
provider	string	No	ollama	LLM provider (default: ollama)
model	string	No	llama3	Model name (default: llama3)
api_key	string	No		API key
endpoint	string	No		API endpoint URL
claim	string	No		The claim or output to verify
evidence	string	No		Evidence or context to verify against
criteria	string	No		Verification criteria (comma-separated or natural language)
verifier_type	string	No	factual	Type: factual, code_correctness, security, logic, consistency, custom (default: factual)
output_format	string	No	detailed	Output: pass_fail, score, detailed, json (default: detailed)
rubric	string	No		Custom scoring rubric for verification (optional)

---

wait

Pause the workflow for a duration (delay/sleep), then pass input through

• Input: string - input text, passed through unchanged after the wait
• Output: string - the input, unchanged

Parameters

Name	Type	Required	Default	Description
duration	string	Yes		How long to wait (Go duration format: 500ms, 10s, 2m, 1h; max 1h — use aflare schedule for longer gaps)

---

xai

Call xAI Grok LLM API

• Input: string - user message content
• Output: string - AI response content

Parameters

Name	Type	Required	Default	Description
model	string	No	grok-4	Model name (default: grok-4)
api_key	string	No		xAI Grok API key (or set XAI_API_KEY env var)
endpoint	string	No	https://api.x.ai/v1	API base URL (default: https://api.x.ai/v1)
system	string	No		System prompt
temperature	string	No		Sampling temperature 0.0-2.0 (default: provider default)
max_tokens	string	No		Max tokens to generate
top_p	string	No		Nucleus sampling probability mass 0.0-1.0
frequency_penalty	string	No		Penalty for repeated tokens -2.0 to 2.0
presence_penalty	string	No		Penalty for new tokens -2.0 to 2.0
stop	string	No		Stop sequences (comma-separated, e.g. '\n,END')
seed	string	No		Random seed for deterministic sampling (int)
response_format	string	No		Structured output: 'json_object' or 'json_schema:<schema_json>'
tools	string	No		JSON array of tool definitions for function calling
tool_choice	string	No		Tool selection: 'none', 'auto', or JSON object
user	string	No		End-user identifier for provider-side abuse monitoring

---

xverse

Call XVERSE LLM API

• Input: string - user message content
• Output: string - AI response content

Parameters

Name	Type	Required	Default	Description
model	string	No	XVERSE-7B-Chat	Model name (default: XVERSE-7B-Chat)
api_key	string	No		XVERSE API key (or set XVERSE_API_KEY env var)
endpoint	string	No	https://api.xverse.cn/v1	API base URL (default: https://api.xverse.cn/v1)
system	string	No		System prompt
temperature	string	No		Sampling temperature 0.0-2.0 (default: provider default)
max_tokens	string	No		Max tokens to generate
top_p	string	No		Nucleus sampling probability mass 0.0-1.0
frequency_penalty	string	No		Penalty for repeated tokens -2.0 to 2.0
presence_penalty	string	No		Penalty for new tokens -2.0 to 2.0
stop	string	No		Stop sequences (comma-separated, e.g. '\n,END')
seed	string	No		Random seed for deterministic sampling (int)
response_format	string	No		Structured output: 'json_object' or 'json_schema:<schema_json>'
tools	string	No		JSON array of tool definitions for function calling
tool_choice	string	No		Tool selection: 'none', 'auto', or JSON object
user	string	No		End-user identifier for provider-side abuse monitoring

---

yi

Call Yi LLM API

• Input: string - user message content
• Output: string - AI response content

Parameters

Name	Type	Required	Default	Description
model	string	No	yi-lightning	Model name (default: yi-lightning)
api_key	string	No		Yi API key (or set YI_API_KEY env var)
endpoint	string	No	https://api.lingyiwanwu.com/v1	API base URL (default: https://api.lingyiwanwu.com/v1)
system	string	No		System prompt
temperature	string	No		Sampling temperature 0.0-2.0 (default: provider default)
max_tokens	string	No		Max tokens to generate
top_p	string	No		Nucleus sampling probability mass 0.0-1.0
frequency_penalty	string	No		Penalty for repeated tokens -2.0 to 2.0
presence_penalty	string	No		Penalty for new tokens -2.0 to 2.0
stop	string	No		Stop sequences (comma-separated, e.g. '\n,END')
seed	string	No		Random seed for deterministic sampling (int)
response_format	string	No		Structured output: 'json_object' or 'json_schema:<schema_json>'
tools	string	No		JSON array of tool definitions for function calling
tool_choice	string	No		Tool selection: 'none', 'auto', or JSON object
user	string	No		End-user identifier for provider-side abuse monitoring

---