devchain-cli

Devchain

AI agent orchestration platform for software development teams

Homepage · Quick Start Guides · What's New in v0.12.0

Devchain runs your AI coding agents as coordinated teams — each with their own terminal session, task queue, and chat. Group agents under team leads, assign epics, track progress on a visual board, and let teams scale themselves to match the workload. Supports Claude Code, Codex, Gemini CLI, and OpenCode out of the box.

Features

Agent Teams

Group agents into named teams with team leads that do real management. The Builders team grows itself with the workload and picks the right model for each task — cheaper models for routine changes, top-tier models for harder work. The Planning team supports parallel planning: add multiple Architects on different models and the Brainstormer gathers independent framings from each before drafting the master plan. Choose your allowed providers per team when you create the project.

Session Reader

Full transcript viewer for active agent sessions, built into the Chat page. See every tool call, thinking block, and response with real-time token usage, cost tracking, and compaction events. Supports Claude Code and Codex transcripts with AI turn grouping, collapsible cards, IQR-based token hotspot detection, and keyboard navigation.

Context Tracking

Visual progress bars show each agent's context window usage in real time. Hover for exact token counts (e.g. "49% used — 98k of 200k"). An inline session summary bar in each terminal shows the active model, running cost, context percentage, and compaction count at a glance.

Provider Model Override

Change any agent's provider and model on the fly from the context menu — no template editing required. Model overrides are per-agent and persist across restarts, layering on top of template defaults.

Worktrees

Spin up isolated agent environments on dedicated git branches. Each worktree gets its own agent team, terminals, and chat — run multiple features in parallel and merge when ready. Worktrees run as Docker containers or local processes with full branch isolation.

Container Isolation

Worktree containers are provisioned automatically from the official Devchain image on GHCR. Each container has Claude Code, Codex, and Gemini CLI pre-installed, runs as a non-root user, and shares your git identity for correct commit attribution.

Skills

Browse and sync AI agent skills from community sources (Anthropic, OpenAI, Vercel, and more). Enable or disable skills per project and expose them to agents via MCP tools.

Visual Workflow Board

Kanban-style epic management with drag-and-drop, list view, board context menu, and URL-based filtering. Agents pick up tasks from the board and update status in real time.

Terminal Sessions

Real terminal streaming via tmux and WebSocket. Each agent gets its own session with scrollback, resize support, and inline access from the chat panel.

Code Review

Live pre-commit diff viewer with agent integration, inline comments, @mentions, comment threading, and VS Code-style file navigation.

Multi-Provider

Works with Claude Code, OpenAI Codex, Google Gemini CLI, GLM models, and OpenCode. Switch providers per agent or switch the whole team preset with a single click.

MCP Integration

Full Model Context Protocol support. Agents get access to epics, chat, skills, reviews, and more through MCP tools — auto-configured before each session.

Local-First

All data stored in a local SQLite database. No cloud account required, no data leaving your machine.

Requirements

• Node.js >= 20
• tmux — required for terminal sessions
  • macOS: brew install tmux
  • Ubuntu/Debian: sudo apt install tmux
• Docker — optional, required for container-isolated worktrees
• AI Provider — at least one of:
  • claude CLI
  • codex CLI
  • gemini CLI
  • opencode CLI

Installation

npm install -g devchain-cli

Quick Start

# Start Devchain — opens browser automatically
devchain start

# Start with a specific project
devchain start --project /path/to/your/project

# Run in foreground with logs
devchain start --foreground

# Bind to all interfaces for VM/remote browser access
devchain start --host 0.0.0.0

# Stop the server
devchain stop

On first run, import a template from the project page to provision your agent team. Three templates are included:

Template	Agents	Best for
teams-dev (recommended)	Planning team (Brainstormer + Architect), Builders team (Epic Manager + Coders), Code Reviewer	New projects — auto-scaling Builders, parallel planning, tier-aware routing
5-agents-dev	Brainstormer, Epic Manager, SubBSM, Coder, Code Reviewer	The classic flat workflow (still bundled and supported)
3-agents-dev	Brainstormer, SubBSM, Coder	Faster iteration with lower token overhead

CLI Options

Option	Description
-p, --port <number>	Port to run on (default: 3000 or next available)
--host <address>	Bind address (default: 127.0.0.1; use 0.0.0.0 for remote/VM access — see docs/setup.md)
-f, --foreground	Run in foreground with visible logs
--no-open	Don't open browser automatically
--db <path>	Custom database directory path
--project <path>	Open with a specific project path

License

Elastic License 2.0 — Free to use. You may not provide this software as a managed service or competing commercial offering.