Agently 4.1.4 Release Notes

语言：English · 中文

Agently 4.1.4 围绕执行所有权、长任务交付、durable context、runtime orchestration、capability control、可观测 model/action execution 完成升级。

核心结果

Agently 4.1.4 将 AgentExecution 收敛为稳定的公开 run surface，并把长任务执行、Workspace evidence、ActionRuntime capabilities、TriggerFlow orchestration 和 runtime observation 放进同一套使用形态：

业务输入
  -> AgentExecution
  -> direct / flat / taskboard strategy
  -> Actions / Skills / Workspace / TaskDAG / TriggerFlow
  -> EvidenceEnvelope + Workspace readback
  -> verifier + host guards
  -> final_response + structured result + RuntimeEvents

关键样例代码

Direct AgentExecution

result = (
    agent
    .input("Summarize the renewal risk and recommend the next action.")
    .output({
        "summary": (str, "short business summary", True),
        "risk_level": (str, "low / medium / high", True),
        "next_action": (str, "recommended next action", True),
    })
    .strategy("direct")
    .get_result()
)

data = result.get_data()
text = result.get_text()
meta = result.get_meta()

Task Strategy With Workspace Evidence

result = (
    agent
    .use_workspace("./.agently/tasks/migration-risk")
    .goal(
        "Prepare a migration risk report.",
        success_criteria=[
            "Cover compatibility, rollout, and rollback risks.",
            "Ground each recommendation in available evidence.",
            "Produce a final artifact that can be read back from Workspace.",
        ],
    )
    .effort("medium")
    .strategy("auto")
    .output({
        "executive_summary": (str, "final summary", True),
        "top_risks": ([str], "material migration risks", True),
        "recommended_plan": (str, "recommended rollout plan", True),
    })
    .get_result()
)

final_text = result.get_text()
task_payload = result.get_data()
task_meta = result.get_meta()

Explicit TaskBoard Delivery

execution = agent.create_task(
    goal="Complete the vendor security questionnaire.",
    success_criteria=[
        "Every required question has an answer.",
        "Each answer is grounded in supplied policy evidence.",
        "The final Markdown file is written and read back from Workspace.",
    ],
    execution="taskboard",
    workspace="./.agently/tasks/security-questionnaire",
)

execution.output({
    "final_file": (str, "Workspace path for the final Markdown file", True),
    "summary": (str, "short completion summary", True),
})

result = execution.get_result()

async for item in result.get_async_generator(type="instant"):
    render_status(item.path, item.value)

answer = await result.async_get_text()
data = await result.async_get_data()

Runtime Guidance During A Task

import asyncio

execution = agent.create_task(
    goal="Prepare the incident handoff.",
    success_criteria=["The handoff reflects the latest operator context."],
    execution="flat",
    workspace="./.agently/tasks/incident-handoff",
)

run_task = asyncio.create_task(execution.async_get_data())

await execution.async_add_guidance(
    "Use the newly uploaded incident note as the primary source.",
    author="operator",
)

data = await run_task
meta = await execution.async_get_meta()
guidance_refs = meta["task_refs"]["workspace_refs"]["guidance"]

Workspace Records And Retrieval

workspace = Agently.create_workspace("./.agently/support-memory")

await workspace.put(
    collection="memory",
    kind="project_note",
    content="Customer prefers staged rollout with rollback checkpoints.",
    tags=["customer", "rollout"],
    source={"type": "operator_note"},
)

context = await workspace.retrieve(
    query="What rollout constraints should the migration report remember?",
    tags=["customer", "rollout"],
    sources=["records", "files"],
    budget={"chars": 12000},
    selection="length",
)

exact_hits = await workspace.grep(
    "rollback",
    filters={"collection": "memory", "kind": "project_note"},
)

Session Memory

from agently.core import Session

workspace = Agently.create_workspace("./.agently/support-memory")

session = Session()
session.use_memory(mode="AgentlyMemory", workspace=workspace)

agent = Agently.create_agent("support-agent").use_workspace(workspace)
agent.activate_session(session_id="support-demo")
agent.activated_session.use_memory(mode="AgentlyMemory")

TriggerFlow Execution

from agently import TriggerFlow, TriggerFlowRuntimeData

flow = TriggerFlow(name="approval-backed-workflow")

async def prepare(data: TriggerFlowRuntimeData):
    await data.async_set_state("ticket_id", data.input["ticket_id"])
    return {"ticket_id": data.input["ticket_id"], "amount": data.input["amount"]}

async def finish(data: TriggerFlowRuntimeData):
    decision = data.input if isinstance(data.input, dict) else {}
    await data.async_set_state("approved", bool(decision.get("approved")))

flow.to(prepare).to(finish)

execution = flow.create_execution(auto_close=False)
await execution.async_start({"ticket_id": "T-100", "amount": 1200})
state = await execution.async_close()

Skills With AgentExecution

result = (
    agent
    .use_workspace("./.agently/tasks/release-readiness")
    .use_skills("release-readiness-reviewer")
    .goal(
        "Review release readiness and produce a go/no-go recommendation.",
        success_criteria=[
            "Check validation evidence.",
            "Identify blocking risks.",
            "Return a structured release decision.",
        ],
    )
    .effort("medium")
    .output({
        "decision": (str, "go / no-go", True),
        "blocking_risks": ([str], "release blocking risks", True),
        "followups": ([str], "required follow-up actions", True),
    })
    .get_result()
)

最终推荐用法

场景	最终推荐用法	主要 API / 表面
普通一次性 Agent run	保持 direct run，并消费 AgentExecutionResult。	agent.input(...).output(...).get_result()；result.get_data()；result.get_text()
多语句 run 配置	创建或持有一个 execution draft，再把 prompt、output、actions、Skills、Workspace、strategy 绑定到同一个 draft。	execution = agent.create_execution()；execution.input(...)；execution.output(...)；execution.get_result()
长任务或证据驱动任务	使用 AgentExecution task strategy，配置 goal、success criteria、effort、Workspace 和 auto strategy。	agent.use_workspace(...).goal(..., success_criteria=[...]).effort("medium").strategy("auto").get_result()
显式策略控制	direct 用于普通 request/action execution；flat 用于线性 bounded task work；taskboard 用于 board/dependency coordination。	execution.strategy("direct")；execution.strategy("flat")；execution.strategy("taskboard")
面向用户的最终文本	从 result text facade 读取 task-strategy final text。	result.get_text()；await result.async_get_text()
结构化任务状态	从结构化 result/meta 读取 task status、artifact status、task refs、completion notes、diagnostics。	result.get_data()；result.get_meta()；result.task_refs
Durable records	通过 Workspace 写入 durable records。	workspace.put(collection=..., kind=..., content=..., tags=[...])
Model-hot retrieval context	用 Workspace intelligent retrieval 获取将进入 model request 或 AgentTask work unit 的 records/files。	await workspace.retrieve(query=..., sources=["records", "files"], budget={"chars": ...})
确定性精确搜索	使用 deterministic grep surfaces 做低成本精确查询和诊断。	await workspace.grep(...)；await workspace.grep_files(...)
Session memory	将 Session memory 绑定到 Workspace，并使用内置 memory plugin 管理 global/session memory records。	session.use_memory(mode="AgentlyMemory", workspace=workspace)；agent.activate_session(...)
Workspace file work	文件 read/search/edit/write 保持在 Workspace file actions 内。	agent.enable_coding_agent_actions(...)；Workspace file IO handlers
Shell 与本地命令	Shell 用于 tests、builds、git inspection、bounded diagnostics。	agent.enable_shell(...)；bounded stdout/stderr artifacts
External Actions	显式挂载 actions，并让 ActionRuntime 拥有 planning、dispatch、policy、artifacts、observations。	agent.use_actions(...)；ActionRuntime records；Action artifact refs
Execution resources	将 runtime capabilities 绑定为 ExecutionResources。	ExecutionResource；内置 ACP、Bash、browser、Docker、MCP、Node.js、Python、SQLite providers
Human-in-the-loop work	使用 ExecutionExchange 与 PolicyApproval-backed wait/approval surfaces。	ExecutionExchange；PolicyApproval；console / host-callback exchange providers
Skills usage	通过 AgentExecution/Agent APIs 选择 Skills，并让 SkillsExecutor 构建 context packs 和 capability plans。	agent.use_skills(...)；Skills context packs；Skills capability policy
Dynamic DAG work	使用 TaskDAG 处理 acyclic dynamic planning 和 execution。默认 TaskDAGExecutor.async_run(...) 直接编译到 TriggerFlow；只有需要 block-graph evidence/result mapping 时才显式选择 Blocks。	TaskDAGExecutor.async_run(...)；可选 TaskDAGExecutor.compile_blocks(...) / async_run_blocks(...)
Workflow orchestration	使用 TriggerFlow 处理显式 branching、waiting、pause/resume、runtime streams、durable workflow execution。	Agently.create_trigger_flow(...)；TriggerFlow(...)；flow.create_execution(...)
Runtime streams	delta 用于面向用户文本；instant / structured events 用于 UI state 和 diagnostics。	get_async_generator(type="delta")；get_async_generator(type="instant")；RuntimeEvents
DevTools observation	通过 DevTools 观察 AgentExecution、model requests、actions、TaskBoard progress、exchanges、telemetry。	agently-devtools >=0.1.10,<0.2.0；RuntimeEvent / ObservationEvent bridge

最终升级矩阵

领域	4.1.4 最终升级	最终推荐用法
AgentExecution ownership	AgentExecution 拥有一次 Agent run：prompt state、action execution、task strategy、process stream、result wrapper、run metadata。	使用 AgentExecution 作为 prompt、action、Skill、task、stream、result consumption 的公开 run surface。
Strategy selection	Execution strategy 收敛为 auto、direct、flat、taskboard。	普通工作使用 auto 或 direct；线性 bounded task work 使用 flat；board/dependency coordination 使用 taskboard。
Direct route	Direct execution 保持普通 model-request 和 ActionLoop run 的轻量路径。	用 direct route 处理短 request/response work 和简单 ActionLoop tasks。
Flat route	Flat execution 共享 AgentTask substrate，并可先将 remaining work 交给下一个 work unit，再进入 final verification。	用 Flat 处理需要 evidence、readback、final verification 但不需要 board scheduling 的顺序长任务。
TaskBoard route	TaskBoard execution 共享 AgentTask foundations，并增加 board state、dependency state、patching、continuation、finalization、bounded projection。	用 TaskBoard 处理多部分 deliverables、依赖密集工作、fan-out/fan-in work 和长制品。
Result text	task-strategy results 暴露 final_response；get_text() 与 async_get_text() 优先返回该 final response。	用 result text facades 获取面向用户的最终答案。
Result payloads	execution result payloads 暴露 terminal status、artifact status、final result data、task refs、completion notes、diagnostics。	用 structured result/meta data 驱动应用状态、审计和 UI detail panels。
Streams	AgentExecution streams 暴露 process events、instant items、delta text、retry boundaries、exchange state、action observations、terminal summaries。	从 delta 渲染用户文本；从 instant 或 RuntimeEvents 渲染结构化 UI state。
Structured request completion	AgentExecution 会投影 provisional instant fields，但其持有的 ModelRequest 会继续到自然 parsing、validation、usage/meta 与 request.completed。	instant 只用于 UI 或可取消/幂等准备；AgentTask evidence 与业务决策使用最终 parsed data。
Runtime context	runtime context 保留为 diagnostics；model-hot task prompts 不把具体 runtime timestamps 写入生成制品。	将业务日期放入 caller input 或 source evidence。
Incremental acceptance	TaskBoard acceptance 携带 dirty/cache markers、card/evidence ids、verdict fingerprints、verification refs、counters、progress percent。	用 acceptance metadata 驱动 task status、board UI 和 verification efficiency。
Verifier reuse	TaskBoard final verification 可复用未变化的 green verifier verdict，并将 dirty verifier input 限定到受影响 acceptance items。	让 TaskBoard 只验证变化的 acceptance areas，同时保留 final verifier authority。
Setbacks	TaskBoard cards 可用 setback 表示可恢复的 readback、repair、patch、continuation failure。	将 setback 渲染为可恢复 task state，并继续执行已排程 recovery work。
Final verification	final verification 接收 pinned evidence ids、normalized verifier evidence、artifact refs、readback facts、acceptance locators、completion notes、unresolved-criteria metadata。	使用 verifier output 与 host guards 作为最终 task acceptance path。
Runtime guidance	active task-strategy execution 可接受 runtime guidance，并在下一个安全边界前存为 Workspace guidance records。	用 add_guidance(...) / async_add_guidance(...) 为 active task runs 追加 operator context。
Evidence ledger	EvidenceEnvelope.evidence_items 是 Flat synthesis、TaskBoard synthesis、verifier prompts、host guards、artifact locators 的 canonical grounding ledger。	源依据重要时，通过 structured outputs 将 output claims 绑定到 evidence ids。
Evidence binding	host guards 将 evidence handles、paths、records、URLs、artifacts、action ids、action-call ids、provenance aliases 统一到 canonical ledger ids。	在 structured result fields 中使用 visible evidence handles 或 canonical ids。
Artifact delivery	Workspace artifact delivery 记录 write facts、readback facts、SHA-256、byte counts、previews、file refs、manifests、targeted readbacks、acceptance locators。	通过 Workspace files 与 readback-backed artifact refs 交付长制品。
Binding repair	binding repair 面向 unresolved evidence bindings 定向修复，不重新生成完整 deliverables。	用 targeted repair 修复 source-binding failures。
Workspace foundation	Workspace 成为 records、files、evidence links、checkpoints、runtime event storage、artifact refs、file policy metadata、retention anchors、leases、backend capability reporting 的 durable boundary。	将一个 Workspace 绑定到需要共享 durable context 的 Agents、TriggerFlow executions、service workers。
Local Workspace backend	local backend 使用 filesystem storage 与 SQLite records、WAL、busy timeout、scope indexes、lineage-aware file roots、scoped prune。	用 local Workspace 支撑开发、本地 durable state、examples、filesystem-backed artifacts。
Workspace writes	workspace.put(...) 是 canonical record-write API，并支持 content=... 与 profile handlers。	用 workspace.put(...) 写 records。
Workspace providers	Workspace backend providers 可通过 Workspace provider seam 注册和选择。	通过 Workspace provider registration 注册 custom backends，并在 Agent 或 execution boundary 绑定。
Workspace file IO	Workspace file IO 拥有 path containment、file refs、deterministic file info、handler dispatch、text read/write、optional export handlers、diagnostics。	将 file IO、export、file-action roots 保持在 Workspace 内。
Intelligent retrieval	workspace.retrieve(...) 为 records/files 提供共享 intelligent retrieval，包含 keyword/tag candidates、optional vector/hybrid candidates、rerank、refill、budgeted packaging。	records/files 要作为 model context 或 AgentTask evidence 时使用 retrieve(...)。
Deterministic search	workspace.grep(...) 和 workspace.grep_files(...) 提供 records/files 的 deterministic exact search。	用 grep(...) / grep_files(...) 做精确查询、调试和诊断。
Workspace store providers	Workspace 将 DBStoreProvider、EmbeddingProvider、VectorStoreProvider 拆开：默认 DB store 是 SQLite，vector_store_provider="auto" 会在 Chroma 可用时选择 Chroma，否则降级到 SQLite vector table。	通过 db_store_provider 接入 record DB adapter，通过 embedding_provider 接入向量化，通过 vector_store_provider 独立选择向量存储。能力较低的 DB store 保持同一协议面，对不支持的高级能力返回空值或缺省值。
Session memory	SessionMemory 成为 plugin protocol；内置 AgentlyMemory 将 global/session memory 存储为 Workspace records。	用 AgentlyMemory 实现 Workspace-backed Session memory 和 scoped recall。
Blocks	Blocks 将 AgentTask ExecutionPlan / PlanBlock work 降低为 TriggerFlow-backed ExecutionBlockGraph，并为已校验 TaskDAG nodes 提供显式可选 carrier。	AgentTask 用 Blocks 承载 bounded work unit；TaskDAG 仅在调用方需要 block lifecycle evidence 或 result adapter 时显式调用 compile_blocks(...) / async_run_blocks(...)。
TaskDAG	TaskDAG 拥有 acyclic dynamic planning、validation、resolver binding、execution、retry metadata、result adaptation、evidence mapping。	用 TaskDAG 直接处理显式 DAG-shaped automation 和 dynamic planning。
TriggerFlow	TriggerFlow 增加 durable snapshots、pause/continue、interrupt/resume ledgers、RuntimeEvent persistence、exchange metadata、compaction policy、load inspection、resource requirements、idempotent resume ids。	用 TriggerFlow 处理需要显式 orchestration、waits、resume、runtime streams、durable execution state 的 workflow。
ExecutionExchange	ExecutionExchange 提供 approvals、decisions、control messages、clarifications、guidance、acknowledgments 的 exchange manager。	用 exchange providers 和 PolicyApproval-backed wait surfaces 处理 human-in-the-loop flows。
ActionRuntime	ActionRuntime 拆分 action planning、dispatch、policy approval、execution、artifact management、resource binding、observation records。	显式挂载 actions，并从 ActionRuntime records 检查 execution facts。
ExecutionResource	ExecutionResource 拥有 ACP、Bash、browser、Docker、MCP、Node.js、Python、SQLite runtimes 的 provider-backed runtime binding。	将 runtime capabilities 作为 resources 绑定，避免在业务代码中嵌入 provider mechanics。
ACP and MCP	ACP 同时是 Action 和 ExecutionResource(kind="acp")；MCP-declared artifacts 通过 Action artifact refs 和 AgentTask evidence handoff 流转。	在 capability boundaries 启用 ACP 或 MCP，并通过 evidence/readback paths 消费 artifact refs。
Workspace file actions	coding-agent Workspace actions 暴露 file read、glob、grep、edit、unified-diff patch、stale-guarded write。	repository/file tasks 使用 Workspace file actions；tests、builds、diagnostics 使用 shell。
Browse and Search	Browse 与 Search actions 使用 policy-controlled execution、fallback behavior、bounded outputs、explicit diagnostics。	将 Browse/Search 作为 mounted capabilities 使用，并消费 bounded output records。
SkillsExecutor	SkillsExecutor 记录 capability needs、构建 context packs、发现/激活 capabilities，并暴露 TaskDAG resolver support。	用 agent.use_skills(...) 和 Skills context packs 支撑 Skill-guided AgentExecution work。
Skills diagnostics	Direct Skills execution 发出 structured abort diagnostics；react/staged strategies 发出 budget-exhausted diagnostics。	将 Skills diagnostics 暴露到 host logs、streams 或 DevTools views。
Model requesters	Model requester providers 模块化为 credential、handler、request-builder、response-adapter、transport、type、plugin modules。	通过 model keys、provider settings、requester plugins 配置 model providers。
Model routing	Model routing 支持 layered model keys、provider fallback、API key pools、request-time key selection、provider-error retry policies。	用 model keys 和 pool settings 处理 provider fallback 与 key rotation。
Model liveness	Model response materialization 为 first event、stream、non-streaming response、materialization stages 提供 liveness deadlines。	用 liveness diagnostics 定位 stalled provider stages。
Stream retry status	ModelRequestResult 暴露 $status records 和 plain delta retry replay markers。	structured stream state 消费 $status；plain text replay boundaries 消费 retry markers。
Telemetry	Model request telemetry 记录 response ids、attempts、run ids、provider/model data、request URLs、duration、usage summaries、side-channel facts、errors、estimated input/output lengths。	将 telemetry 提供给 DevTools 和 host diagnostics。
Structured output	output defaults 由 settings 拥有；已发布 parsers 包含 xml_field、hybrid、JSON、yaml_literal、flat_markdown；required fields 强制 meaningful values。	model-owned structured decisions 使用 .output(...) 和 Agently output control。
Image input	VLM helpers 可从 local files、URLs、bytes 或 structured image payloads 构建 rich image input。	VLM input 使用 agent.image(...) / request image helpers。
RuntimeEvent	RuntimeEvent 是 core runtime event record；EventCenter dispatches RuntimeEvents，并支持 delivery policy、coalescing、background reclaim。	使用 RuntimeEvents 作为统一 observation feed。
DevTools	DevTools 消费 AgentExecution streams、model status、task progress、action observations、exchange states、retry status、terminal summaries、telemetry。	Agently 4.1.4 搭配 agently-devtools >=0.1.10,<0.2.0。
Public typing	包内发布 agently/py.typed，并扩展 facades、protocols、TypedDicts、data contracts、callbacks、stream handlers、result wrappers、Workspace、ExecutionExchange、TaskBoard helpers 的 typing。	对安装后的 package 使用 pyright/Pylance-compatible tooling。
Docs and examples	文档和 examples 覆盖 AgentExecution strategy、Workspace retrieval、Session memory、Action Runtime、ExecutionResource、TriggerFlow lifecycle、Skills execution、DevTools observation、structured output、release workflows。	新 examples 从 4.1.4 AgentExecution、Workspace、TriggerFlow、Skills、ActionRuntime surfaces 开始。