AI agent orchestration

Orchestrate agent CLIs like production workflows.

Dagu gives AI agent commands the operational wrapper they need: scheduling, dependency order, retries, logs, artifacts, and human checkpoints.

Agent orchestration as YAML
name: agent-release-notes
schedule: "0 9 * * 1"

steps:
  - id: collect_changes
    run: git log --since="7 days ago" --oneline
    output: GIT_LOG

  - id: draft_notes
    action: harness.run
    with:
      prompt: |
        Draft release notes from this git context:
        ${GIT_LOG}
    depends: [collect_changes]

  - id: human_review
    run: ./scripts/request-approval.sh
    depends: [draft_notes]

Works with agent CLIs instead of replacing them

Human review steps fit naturally in the graph

Logs and artifacts stay under your control

Model and provider choices remain portable

At a glance

Dagu's role in an agent stack

Agent model
Dagu

Bring any CLI or API-backed agent.

Typical alternative

Vendor platforms often tie orchestration to their own model surface.

Operations
Dagu

Schedules, retries, logs, artifacts, approvals, and reruns.

Typical alternative

Prompt-only automation leaves these controls to custom glue code.

Ownership
Dagu

Workflow definitions, logs, and runtime live in your environment.

Typical alternative

Hosted harnesses can make workflow state and pricing harder to control.

In depth

Where each tool fits

01

Keep the harness under your control

Agent workflows are more than one prompt. They collect context, call tools, validate output, request approval, and publish artifacts. Dagu keeps that harness in your own repo and runtime.

  • Treat agent calls as workflow steps with dependencies and retries.
  • Capture stdout, stderr, artifacts, and run history for review.
  • Add validation and approval steps before generated output reaches users.
02

Use the agent tools your team already trusts

Dagu is not an agent framework. It runs the command-line tools and scripts your team already uses, which keeps the orchestration layer portable.

  • Call Claude, Codex, Gemini, OpenCode, Aider, or internal tools.
  • Switch providers by editing commands and environment variables.
  • Avoid locking production workflows to one vendor harness.
03

Schedule recurring agent work

Release notes, triage summaries, cleanup passes, reporting, and QA checks become scheduled workflows instead of manual prompts.

  • Run agent workflows on cron schedules.
  • Fan out over repositories, teams, or environments.
  • Record every run so humans can inspect what happened.

FAQ

Practical questions before adopting Dagu

Is Dagu an AI agent framework?

No. Dagu is the command workflow layer underneath agent work. It schedules and observes commands, so you can use the agent CLI, model, and prompt system that fits each job.

How do human approvals work?

Approval can be modeled as its own workflow step. Teams commonly add review scripts, notification steps, or manual gates before publishing agent-generated output.

Can Dagu run multiple agent providers?

Yes. Because steps are commands, a workflow can call different CLIs or scripts in different steps and use fallback logic around them.

Next step

Start with one workflow.

Install Dagu, move one fragile script or agent task into YAML, and decide from a real run history.