Octogent
A local dashboard for many Claude Code sessions: each job gets its own context and todo, agents spawn child agents and message each other
Install
pnpm install
pnpm build
npm install -g .
octogentRun inside a cloned repository; the package is not on npm yet.
This is third-party code. Review the repository files before installing.
What it does
Octogent splits work into tentacles: folders under .octogent/tentacles holding CONTEXT.md, todo.md and notes for one part of the codebase. From todo.md items you can launch individual Claude Code agents or a swarm that works through the list in parallel. One agent can assign work to others and receive short completion and blocker messages. Terminals run in the shared workspace or in separate git worktrees, and a local API and web UI show their state through Claude hooks.
Who it is for. For developers running many Claude Code sessions at once across different parts of a project.
Good fit when
- Docs, API, database and frontend work run in parallel
- You need to split a large job into items and hand them to several agents
- You want to see and control each worker agent, not only a subagent's result
Not a fit when
- You work with a single agent session
- You use an agent other than Claude Code; docs cover only it
- You need a stable npm install; the package is not published yet
Example request
Assign the open items in the api tentacle's todo.md to separate agents and tell me when all are doneLimitations
A personal, experimental project; not published to npm and installs only from a clone. Requires Node.js 22, claude, git, plus gh and curl for pull request features. Terminal sessions do not survive an API restart and are capped at 32 by default.
How to disable. Stop terminals with octogent terminal stop and prune, remove the global package with npm uninstall -g octogent, and delete .octogent in the project and ~/.octogent.
Security check
- Launches many agents that change code in parallel
- Runs a local API and installs Claude Code hooks
README in short
The README describes the pain of juggling many Claude Code sessions and the tentacle model with context and task lists. It explains Claude Code coordinating other agents and how context, execution and worktree isolation are separated. Install is from source with pnpm for now, with requirements and persisted state listed. MIT licensed.
FAQ
How is this different from Claude Code subagents?
Each worker runs in its own terminal, so you can see and control what it does directly.
Where is state stored?
Context and todos in the project's .octogent, runtime state and transcripts in ~/.octogent/projects/<id>/state.
Related
Open-source personal AI assistant on your own machine: answers in Telegram, Slack, Discord and WhatsApp, extended with skills and plugins
A self-improving agent from Nous Research with a TUI, messaging gateway, cron jobs and skills it writes itself
An open source coding agent for the terminal and desktop with build and plan modes
Open prompt library with a Claude Code plugin, MCP server and CLI: search, fetch and improve prompts and skills from an agent