Skip to main content
Runtime config lives at ~/.operator/operator.yaml.

​
Defaults

defaults:
  models:
    - "anthropic/claude-opus-4-6"
    - "openai/gpt-5.3-codex"
  max_iterations: 25
  context_ratio: 0.5
  max_output_tokens: null
  env_file: "~/.env"
FieldDescription
modelsFallback chain — if the first model errors, the next is tried. Always use list format.
max_iterationsMaximum tool-call loops per agent run.
context_ratioFraction of context window to use for conversation history.
max_output_tokensCap output length per LLM call. null uses each model’s max.
env_filePath to env file for API keys. Supports ~ expansion.

​
Agents

agents:
  operator:
    models:
      - "anthropic/claude-sonnet-4-6"
    transport:
      type: slack
      bot_token_env: SLACK_BOT_TOKEN
      app_token_env: SLACK_APP_TOKEN
Each agent can override models, max_iterations, context_ratio, and max_output_tokens from defaults. Agents without a transport block are available for jobs and sub-agent spawning but have no chat interface. See Agents for details on agent prompts and workspaces.

​
Memory

memory:
  embed_model: "openai/text-embedding-3-small"
  embed_dimensions: 1536
  max_memories: 10000
  inject_top_k: 5
  inject_min_relevance: 0.1
  harvester:
    enabled: true
    schedule: "*/30 * * * *"
    model: "openai/gpt-4.1-mini"
  cleaner:
    enabled: true
    schedule: "0 3 * * *"
    model: "anthropic/claude-haiku-4-5"
FieldDescription
embed_modelRequired when any memory service is enabled.
embed_dimensionsEmbedding vector size.
max_memoriesPer-scope soft cap.
inject_top_kMemories injected per message.
inject_min_relevanceCosine similarity threshold.

​
Harvester

Extracts facts from conversations using an LLM and stores them as vector embeddings in SQLite (via sqlite-vec). When enabled, schedule and model are required.

​
Cleaner

Deduplicates, merges, and tidies stored memories by sending them through an LLM. When enabled, schedule and model are required. See Memory for more on scopes, pinning, and retrieval.

​
Environment Variables

API keys are resolved from the environment, loaded via env_file: 
# LLM providers
ANTHROPIC_API_KEY=sk-ant-...
OPENAI_API_KEY=sk-...

# Slack (per agent)
SLACK_BOT_TOKEN=xoxb-...
SLACK_APP_TOKEN=xapp-...
All model names use LiteLLM format (provider/model-name).

​
Settings

settings:
  show_usage: true
FieldDescription
show_usageAppend token usage stats as a reply after each agent response. Defaults to false.