Configuration reference⚓︎
ContextAgent uses YAML for pipeline definitions. This reference outlines the core sections, defaults, and accepted values.
Top-level schema⚓︎
version: 1
defaults:
model: gpt-4.1-mini
temperature: 0.3
context_store:
provider: memory
artifact_store:
provider: disk
path: outputs/artifacts
agents:
planner:
profile: research/planner@v1
pipelines:
research:
entrypoint: pipelines.research.ResearchPipeline
agents⚓︎
| Field | Type | Description |
|---|---|---|
profile |
str |
Profile bundle to hydrate prompt + defaults |
model |
str |
Override the model for this agent |
planner |
bool |
Enable planning phase before execution |
tools.allow |
list[str] |
Tool scopes the agent can call |
Example:
agents:
writer:
profile: research/writer@v2
model: claude-3-sonnet
planner: false
tools:
allow: [github_insights, web_search]
tools⚓︎
| Field | Type | Description |
|---|---|---|
path |
str |
Python import path to the tool class |
timeout |
int |
Seconds before the tool call aborts |
scopes |
list[str] |
Access scopes required by agents |
secrets |
list[str] |
Environment variables that must be set |
pipelines⚓︎
| Field | Type | Description |
|---|---|---|
entrypoint |
str |
Import path to the pipeline class |
inputs |
dict |
Pydantic schema metadata for inputs |
schedule |
str |
Optional cron expression for recurring runs |
Observability⚓︎
telemetry:
exporter: otlp
endpoint: ${OTLP_ENDPOINT}
logging:
level: info
redactions:
fields:
- api_key
- password
Note
Observability settings cascade to both agents and tools. Override per-component if you need granular control.
Environments⚓︎
Use environment overlays to manage staging vs production:
environments:
staging:
defaults:
model: gpt-4o-mini
production:
defaults:
model: gpt-4.1
context_store:
provider: redis
url: ${REDIS_URL}
Selecting an environment⚓︎
Validation⚓︎
Validate configs with the built-in checker: