First Tree

First Tree is an open-source, Apache 2.0-licensed workspace built by Agent Team Foundation that lets engineering teams run multiple AI coding agents in parallel from a single hub. The CLI and core platform are freely available on GitHub, while a hosted cloud hub adds team collaboration features. The project is written in TypeScript and actively maintained, with the latest release at v0.5.18 as of July 2026.

What It Is

First Tree is an agent orchestration layer — not another AI agent itself, but the coordination surface that connects agents like Claude Code, Codex, Cursor, OpenCode, and Kimi into a unified team workflow. Its central concept is the Context Tree: a Git-native, team-maintained memory of decisions, ownership, responsibilities, constraints, and prior work. Agents read this shared context before starting any task, and useful outcomes can flow back into it after work is done, creating a persistent human-agent work loop.

How the Context Loop Works

The core architecture connects five pieces:

• Context Tree — a structured, file-based team memory stored in Git, covering goals, product positioning, system design, and team practices
• Web workspace — the daily surface for chats, agent management, GitHub integration, and context-backed task tracking
• CLI + daemon — signs a local computer into the hub and keeps local agents connected and routed
• Agent runtime — runs agents on the user's machine and routes messages through First Tree
• GitHub integration — connects code work, pull requests, and reviews back to the workspace

The homepage describes the loop as: user intent → read team context → context-aware agent work → human review/control → durable outcome → updated team context.

Parallel Agent Workflow

First Tree is designed for two primary work modes:

• Focused copilot work — deep collaboration with a single agent in a persistent work stream for coding, design, writing, or research
• Parallel review work — multiple agents move tasks forward simultaneously, looping in humans only when a decision, blocker, approval, or review is needed

Each agent can be given a distinct role, model, prompt, and skill set. Agents coordinate in shared chats, and humans can join any agent chat at any time. GitHub issues can be automatically triaged and assigned to agents, which open pull requests rather than merging directly — access is scoped to only the repos and branches explicitly granted.

Open-Source Deployment Model

The CLI and core platform are released under Apache 2.0, allowing self-hosted deployments with any LLM provider. The hosted First Tree Hub at cloud.first-tree.ai provides a free tier, with paid plans adding seats and analytics according to the FAQ on the homepage. The macOS/Linux installer bundles Node.js so new users do not need a separate Node installation. Self-hosted deployments use Docker Compose for the development PostgreSQL database and a Fastify API server.

Update: v0.5.18 — BYO Context, GitHub Task Agents, Complete Team Skills

The GitHub repository shows the latest release as v0.5.18, published July 30, 2026, titled "BYO Context, GitHub task agents, complete Team Skills." The repository was created in March 2026 and has been actively pushed to as recently as August 2026, indicating rapid early development. The project also ships a companion open-source security scanner (launch-readiness-scan) powered by GPT-5.6 that audits public repos across eight categories including secrets, auth, input safety, and deploy config.

Team and Background

The team page lists six members with backgrounds from Google, Alibaba (via Ledong acquisition), Kuaishou, Baidu, ByteDance, Xigua Video, Lark, and UC Berkeley. CEO Chao Liu was previously a Google PM and founder of Ledong. The organization is based in San Francisco and operates under the Agent Team Foundation GitHub organization.