Run Agents.
Get Patches.
Ship PRs.
Run AI coding agents in a sandbox, produce standardized artifacts, and wire them into CI to turn issues into PRs.
Holon is a headless, containerized agent runner built for automation, review, and shipping — not demos.
Why Holon?
AI coding agents are good at generating code, but shipping software still requires repeatability, review, and validation.
Explicit Executions
Turn vague agent text runs into explicit, reproducible executions against a snapshot of your codebase.
Reviewable Artifacts
Turn model output into standard diff.patch files that you can verify, review, and apply.
CI-Friendly Workflows
Turn interactive experiments into deterministic steps in your GitHub Actions pipelines.
How It Works
Holon doesn’t replace your workflow. It slots into it.
Trigger
Issue or CI Workflow
Holon Run
Sandboxed Agent Execution
Artifacts
diff.patch, summary.md
Review & Ship
Apply Patch & Open PR
Standardized Artifacts
Predictable outputs every run.
diff.patch— explicit code changessummary.md— human-readable intent/resultsmanifest.json— structured metadata
Sandbox by Default
Run agents safely and explicitly.
- Execute against a workspace snapshot
- No silent file mutations
- Apply patches only when you choose
CI-First by Design
Made for GitHub Actions.
- Trigger runs from issues or workflows
- Produce PR-ready outputs
- Fit naturally into existing pipelines
Engine-Agnostic
Bring your own agent.
- Claude Code, Codex, OpenHands, etc.
- Stable agent contract
- No model or vendor lock-in
What Holon Is
- A deterministic execution layer for AI agents
- A way to turn agent runs into shippable work
- Infrastructure you can trust in CI
What Holon Is Not
- A chat interface
- A prompt framework
- An autonomous agent that bypasses review
Start running agents like real engineering work.
Holon doesn’t generate code. It ships work.