Qualixar OS is configured via ~/.qualixar-os/config.yaml. The schema is defined with Zod in src/types/common.ts (QosConfigSchema). All fields have defaults; an empty file is valid.
mode: companion
routing: balanced
providers:
local:
type: ollama
endpoint: http://localhost:11434
models:
primary: ollama/llama3
fallback: ollama/llama3
budget:
max_usd: 100
warn_pct: 0.8
execution:
max_output_tokens: 16384
agent_quality: balanced
enable_shell: false
All fields have defaults; an empty file is valid.
| Field | Type | Default | Description |
|---|
mode | companion | power | companion | System mode (affects feature gates) |
routing | quality | balanced | cost | balanced | Model routing strategy |
Each provider is a named entry with these fields:
| Field | Type | Required | Description |
|---|
type | string | Yes | Provider type: ollama, openrouter, anthropic, openai, azure-openai, google, lmstudio, custom |
api_key_env | string | No | Env var name holding the API key |
endpoint | string | No | Base URL (resolved from endpoint_env if not set) |
endpoint_env | string | No | Env var name holding the endpoint URL |
api_version | string | No | API version (Azure) |
deployments | object | No | Model-to-deployment mapping (Azure) |
| Field | Type | Default | Description |
|---|
primary | string | claude-sonnet-4-6 | Default model for tasks |
fallback | string | gpt-4.1-mini | Fallback when primary fails |
judge | string | (optional) | Model for judge assessments |
local | string | (optional) | Preferred local model |
catalog | array | [] | User-defined model entries |
Each catalog entry: { name, provider, deployment?, quality_score (0-1), cost_per_input_token, cost_per_output_token, max_tokens }.
| Field | Type | Default | Description |
|---|
max_usd | number | 100 | Global hard spending limit |
warn_pct | number | 0.8 | Warning threshold (0-1) |
per_task_max | number | (optional) | Per-task spending cap |
| Field | Type | Default | Description |
|---|
container_isolation | boolean | false | Enable container sandboxing |
policy_path | string | (optional) | Custom security policy file |
allowed_paths | string[] | ["./"] | Paths agents can access |
denied_commands | string[] | ["rm -rf", "sudo"] | Additional blocked commands |
| Field | Type | Default | Description |
|---|
enabled | boolean | true | Enable memory system |
auto_invoke | boolean | true | Auto-recall before tasks |
max_ram_mb | number | 50 | Memory RAM budget |
embedding.provider | string | azure | Embedding provider |
embedding.model | string | text-embedding-3-large | Embedding model |
embedding.dimensions | number | 3072 | Vector dimensions |
| Field | Type | Default | Description |
|---|
max_output_tokens | number | 16384 | Max output tokens per agent call (1024-32768) |
agent_quality | balanced | high | maximum | balanced | Agent quality tier |
enable_shell | boolean | false | Allow shell command execution |
Array of MCP tool connector entries:
| Field | Type | Required | Description |
|---|
id | string | Yes | Unique identifier |
name | string | Yes | Display name |
transport | stdio | streamable-http | No | Transport type (default: stdio) |
command | string | No | Command to start (stdio) |
args | string[] | No | Command arguments (stdio) |
url | string | No | Server URL (streamable-http) |
| Field | Type | Default | Description |
|---|
default_dir | string | (empty) | Custom workspace base directory |
dashboard: { enabled: boolean, port: number }channels: { mcp, http, telegram, discord, webhook } -- each with enabled toggleobservability: { otel_endpoint?, log_level } -- OpenTelemetry and loggingdb: { path } -- SQLite database location
Config changes are auto-detected via file watcher. The server emits a config:changed event when the file is modified.