sortie

Your next sprint isn't capped by headcount.

The open-source coding agent orchestrator. Run the coding agents you already use on tasks from your issue tracker - in parallel, on your own machine or server.

Website · Documentation · Discord · Contributing

English | 简体中文

The Problem

Coding agents can fix bugs, update dependencies, add tests, and build features. But running agents you trust across a backlog takes more than launching sessions: isolated workspaces, retries, tracker integration, and cost tracking all need to work together.

Building and maintaining that infrastructure becomes another engineering project.

Sortie is that infrastructure.

Works With

Issue trackers: GitHub Issues, GitLab Issues, Gitea Issues, Linear and Jira.

Coding agents: Claude Code, Copilot, OpenCode, Codex, Kiro and Gemini.

Install

curl -sSL https://get.sortie-ai.com/install.sh | sh

Or via Homebrew: brew install --cask sortie-ai/tap/sortie

Try the local demo with a simulated agent and no external services, or connect your tracker and coding agent.

How It Works

You describe the work in one WORKFLOW.md file: which issues to pick up, which agent to run, and what to tell it.

---
tracker:
  kind: github
  api_key: $GITHUB_TOKEN
  project: acme/billing-api
  query_filter: "label:agent-ready"
  active_states: [todo, in-progress]
  handoff_state: review
  terminal_states: [done]
  
agent:
  kind: claude-code
  max_concurrent_agents: 4
---
  
You are a senior engineer.
  
## {{ .issue.identifier }}: {{ .issue.title }}
  
{{ .issue.description }}

Sortie polls your tracker for matching issues and runs an agent on each one in its own workspace, several at a time. A failed run is retried. Turn on CI and review feedback, and failed checks and review comments go back to the agent to fix.

You can edit WORKFLOW.md while Sortie runs, and it picks up the change without a restart. Run history survives restarts too, so progress and costs are still there afterwards.

Follow along in the built-in dashboard, or connect your own tools: a JSON API with a live stream of changes, health probes, and Prometheus metrics, all on one port.

Sortie runs the same way on a laptop, a VM, or a Kubernetes cluster. It ships as a container image at ghcr.io/sortie-ai/sortie, its health probes plug straight into liveness and readiness checks, and it shuts down cleanly when the pod stops. The Kubernetes guide walks through a full deployment.

Sortie is a single binary. There's no separate database or job queue to deploy, and Sortie itself sends no telemetry.

Documentation

Full configuration reference, CLI usage, and getting started guide: docs.sortie-ai.com

• Guides
• Architecture
• Roadmap
• Code signing policy

Prior Art

Sortie's architecture is informed by OpenAI Symphony, a spec-first orchestration framework with an Elixir reference implementation. Sortie diverges in language (Go for deployment simplicity), persistence (SQLite instead of in-memory state), extensibility (pluggable adapters for any tracker or agent, not hardcoded to Linear and Codex), and completion signaling (orchestrator-managed handoff transitions instead of relying solely on agent-initiated tracker writes).

License

Apache License 2.0