🪨 why use many token when few token do trick — Claude Code skill that cuts 65% of tokens by talking like caveman
Local agent orchestration for Claude Code and Codex
DevChain runs AI coding agent CLIs in real tmux terminals and keeps their work organized in a web app. It gives you teams, epics, chat, and code review, while storing state locally in SQLite and connecting to agents through MCP tools.
Builders who want multiple agent CLIs to plan, code, and review together on one machine.
You can coordinate agent work, watch it run live, and review changes from one local dashboard.
What it does
Coordinated agent teams
Runs Claude Code, Codex, OpenCode, Antigravity, and Copilot as teams that can split up work and switch models per task.
Live tmux sessions
Streams each agent in its own real terminal with full TTY support, so you can watch, scroll back, or take over.
Board-driven workflow
Uses kanban-style epics and sub-epics that agents can pick up and update through MCP tools.
Built-in code review
Shows diffs with inline comments, @mentions, and threaded discussion inside the workflow.
Local state and transcripts
Stores sessions, transcripts, token usage, cost, and compaction events in local SQLite.
Skills and MCP setup
Syncs community skill packs and auto-configures MCP tools before each session.
How to get it
- 1Requirements: Node.js 24 LTS or newer, tmux (brew install tmux / sudo apt install tmux),…
npm install -g devchain-cli devchain start
README
Quick Start · Docs · Mobile App · Releases
DevChain is a local-first orchestrator for AI coding agents. It runs Claude Code, Codex, OpenCode, Antigravity, and GitHub Copilot as coordinated teams — each agent in its own real tmux terminal, sharing a board of epics, a chat, and a code-review flow. You describe the work, the teams plan and build it in parallel, and you review and merge from the web UI — with an end-to-end-encrypted mobile app to keep an eye on everything when you step away.
Key features
- Teams that manage themselves — the Builders team adds Coders when work piles up and picks the right model per task: cheaper models for routine changes, top-tier models for harder work. The Planning team researches plans from several angles in parallel before you approve them.
- Real terminals, live — every agent runs in a tmux session streamed to your browser with full TTY support. Watch it work, scroll back, or take over at any moment.
- A board agents actually use — kanban-style epics and sub-epics with drag-and-drop; agents pick up tasks and update statuses themselves through MCP tools.
- ClickUp and Jira integration — connect your work boards to see assigned tasks in DevChain, import them as Epics with links back to the source, manage statuses and subtasks, and log time without leaving the app.
- Code review built in — live pre-commit diff viewer with inline comments,
@mentions, and threading, wired into the agent workflow. - Session reader and context tracking — full transcript viewer for all five providers, with per-turn token usage, cost tracking, and compaction events; live context-window bars for every agent.
- Skills and MCP — sync community skill packs (Anthropic, OpenAI, Vercel, and more) and expose them to agents; the full MCP toolset is auto-configured before each session.
- Local-first — everything lives in a local SQLite database and runs on your machine, with your provider accounts and keys.
Supported providers
| Provider | CLI | Sessions | Transcripts |
|---|---|---|---|
| Claude Code | claude | ✔ | ✔ |
| Codex | codex | ✔ | ✔ |
| OpenCode | opencode | ✔ | ✔ |
| Antigravity | agy | ✔ | ✔ |
| GitHub Copilot | copilot | ✔ | ✔ |
Model families such as GLM are available through per-provider configs, and you can switch the provider or model of any agent on the fly.
Quick start
Requirements: Node.js 24 LTS or newer, tmux (brew install tmux / sudo apt install tmux), and at least one provider CLI from the table above.
npm install -g devchain-cli
devchain start
The browser opens automatically. Create a project, import a template, and start the Brainstormer session with a description of what you want to build. Run devchain start --help for ports, host binding, and foreground mode; devchain stop shuts the server down.
Native installation
better-sqlite3 v13 ships bundled Node-API artifacts, so mainstream supported installations do not need a C/C++ compiler for SQLite. DevChain's postinstall check opens an in-memory database, executes a query, and closes it; an unusable artifact now fails installation instead of only printing a diagnostic. Set DEVCHAIN_SKIP_POSTINSTALL=1 only when intentionally bypassing that verification.
CI executes the application and SQLite checks on Linux x64 with Node 24, plus a focused Node 26 compatibility lane. Published binaries or image manifests for other platform/architecture combinations prove artifact presence only; they are not runtime-execution evidence for macOS, Windows, or arm64.
| Template | Agents | Best for |
|---|---|---|
teams-dev (recommended) | Planning team (Brainstormer + Architect), Builders team (Epic Manager + Coders), Code Reviewer | Auto-scaling Builders, parallel planning, tier-aware model routing |
How it works
graph LR
Browser["Web UI<br/>(React)"] -->|HTTP + Socket.IO| App["Local App<br/>(NestJS + Fastify)"]
App --> DB[("SQLite<br/>local storage")]
App <-->|tmux / PTY| Sessions["Agent terminal sessions"]
Sessions --- CLIs["Provider CLIs<br/>claude · codex · opencode · agy · copilot"]
CLIs <-->|MCP tools| App
Mobile["Mobile app<br/>(iOS / Android)"] <-->|E2EE relay<br/>sealed data only| App
The Local App serves the web UI and stores all state in a local SQLite database. Agents run as provider CLIs inside tmux sessions and coordinate through DevChain's MCP tools — epics, chat, reviews, skills, and team management. The optional mobile app connects through an end-to-end-encrypted relay: the cloud only forwards sealed data it cannot read.
Mobile app
Monitor and steer your agent teams from your phone, in open beta on iOS (TestFlight) and Android (Play Store). Chat with agents, answer their questions the moment they ask, reassign epics and comment on the board, watch a live terminal viewport, and get push notifications when a session stops or needs you. Reviewing and merging stay on the web — the phone is your mission control on the go. Everything between your PC and phone is end-to-end encrypted.
Resources
- Homepage — product overview and screenshots
- Quick Start Guides — visual step-by-step setup for both templates
- Releases — versioned release notes
- Contributing — development setup and project structure
Community and support
Questions, bug reports, and feature requests are welcome in GitHub Issues. Watch Releases to follow what ships.
Contributing
Development setup, prerequisites, and project structure live in CONTRIBUTING.md. Issues and pull requests are welcome.
License
Free and source-available under the Elastic License 2.0. You may use, copy, and modify DevChain freely; you may not provide it as a managed service or a competing commercial offering.
Files in the repo
- .github
- apps
- packages
- prebuilds
- scripts
- .editorconfig
- .gitattributes
- .gitignore
- .npmrc
- .prettierignore
- CONTRIBUTING.md
- LICENSE
- package.json
- pnpm-lock.yaml
- pnpm-workspace.yaml
- README.md
- turbo.json
Discussion (0)
Ask about usage, or say what you built with itSign in to join the discussion.
No comments yet. Be the first to say what this is good for.
More tools
The best-benchmarked open-source AI memory system. And it's free.
Orca is the ADE for working with a fleet of parallel agents. Run any coding agent with your own subscription. Available on desktop, mobile and remote runtime.

A cross-platform desktop All-in-One assistant for Claude Code, Codex, OpenCode, OpenClaw, Grok Build & Hermes Agent. Only official website: ccswitch.io
Never stop coding. Free MIT AI gateway: one endpoint, 352 providers (150+ free), 1200+ models Kimi, Claude, GPT, Gemini, GLM, DeepSeek, MiniMax. Works with Claude Code, Codex, Cursor, OpenCode, Cline & Copilot. Quota-aware auto-fallback, RTK+Caveman compression saves 15-95% tokens, MCP/A2A, Desktop/PWA. Built by 550+ contributors
Compress tool outputs, logs, files, and RAG chunks before they reach the LLM. 20% fewer tokens for coding agents, 60-95% fewer tokens for JSON, same answers. Library, proxy, MCP server.
