Agents

An agent is the atomic unit of work in a Modulo pipeline – a configured node that performs one step: read a file, call an LLM, write a PR comment, or wait for human review.

Agent configuration

Pipeline config

{
  "name": "draft-pr-description",
  "type": "llm",
  "model_backend": "anthropic",
  "prompt": "Write a PR description for the attached diff.",
  "schema": "pr-description-v1",
  "connector": "github"
}

Each agent has:

Agent configuration
Field	Description
`name`	Unique identifier within the pipeline
`type`	`llm`, `connector`, `transform`, `gate`, `trigger`
`model_backend`	Which model backend to route through (for `llm` type)
`prompt`	System prompt for the agent
`schema`	Schema that validates the agent’s output
`connector`	Connector binding for tool access
`timeout`	Max execution time before cancellation

Prompt versioning

Every prompt change creates a new version. Modulo preserves all versions and allows restoring any previous one. The audit log records which prompt version produced each run output.

Schema assignment

Agents produce typed artifacts. A schema defines the expected shape and constraints. Schemas are versioned independently from agents, so a schema update can affect multiple agents at once.

Gate agents

Gate agents don’t produce output – they evaluate a condition and either pass or block the pipeline:

Pipeline config

{
  "name": "check-approval",
  "type": "gate",
  "condition": "{{ run_context.approved == true }}"
}

Human-only gates

Mark an agent as human_only: true to prevent automated execution. The pipeline pauses at this node until a human manually completes the step through the UI.

Next: Human-in-the-Loop