Agent orchestration
OpenKit
DiscontinuedOpenKit brings coding agents into a shared workspace, enabling orchestration from issue to review.
Start from an issue, give an agent a workspace, and follow the work through to review. OpenKit brings together the tools we needed when we first started running several agents on the same codebase.
What OpenKit does
- One workspace per task
- Create a Git worktree from a Jira, Linear, or GitHub issue. Each agent can work in its own checkout while other tasks run alongside it.
- Development servers that can share a machine
- OpenKit detects the ports your project uses and offsets them per worktree. Runtime hooks cover Node.js and native socket calls, so you can run multiple copies without editing the app.
- Shared tools across agents
- Keep MCP servers, skills, and plugins in one registry and deploy them to the agent tools you use, including Claude Code, Codex, Gemini CLI, Cursor, and VS Code.
- Hooks around the work
- Run checks, prompts, or skills when a worktree is created or removed, before or after an agent task, or when an agent decides a condition you described has been met.
- A place to see what’s happening
- Open agent terminals, follow activity, review a worktree’s diff, and see how much of the machine each workspace is using. OpenKit includes a CLI, local server, web UI, desktop app, and a mobile companion for pairing.
Project reference
Read the CLI command reference or browse the repository documentation for configuration, hooks, integrations, and architecture.
Why development ended
OpenKit gave our agents a shared place to run when they needed more help coordinating work. As they got better at following repository instructions and using the tools directly, we moved those workflows into docs, skills, hooks, CLIs, and routines we could maintain alongside the code.
OpenKit is discontinued, and it’s still a project we’re proud to share. The source is still there for anyone curious about what we built.