First Tree: AI Team Context Workspace | Generated by AI
Here’s what first-tree is:
high-level
It’s an open-source workspace + context layer for human-agent teams. The core idea: instead of every AI agent starting a task cold with an isolated prompt, agents read from — and write back to — a shared, Git-native knowledge base called a Context Tree.
Think of it as team memory that agents and humans maintain together.
architecture
It’s a pnpm monorepo with 5 packages:
| Package | Role |
|---|---|
apps/cli |
first-tree / ftd CLI — login, daemon, agent management, tree operations |
packages/server |
Fastify API server (PostgreSQL/Drizzle/Zod) — the SaaS backend |
packages/client |
SDK + AgentRuntime — wraps WebSocket comms, handler registration, codex/copilot runtime glue |
packages/shared |
Zod schemas + types shared across server/client/cli — source of truth for DTOs |
packages/web |
React 19 + Vite web workspace — the GUI for chats, agents, context tree |
what problem it solves
When an agent — or a human — needs to understand why something was built a certain way, there’s usually nowhere to look. PRs get merged, issues closed, documents decay. The Context Tree is a recursive NODE.md structure (Git repo with markdown files) that holds durable decisions, ownership, constraints, and domain knowledge.
how it works
- Context Tree — a Git repo of markdown files that agents read before acting
- Web workspace — chat surface for agents, team members, GitHub integration
- CLI + daemon — signs a computer in, keeps local agents connected
- Agent runtime — runs agents (Claude Code, Codex, your own) on your machine, routes through First Tree
- GitHub integration — connects PRs, code reviews back into the workspace
The loop is:
user intent -> read team context -> agent works context-aware
-> human review/control -> durable outcome -> updated team context
who made it
From the org name agent-team-foundation/first-tree — built by a team that works deeply with AI agents daily.
how far along
~500 commits on main, CI, Docker compose for local dev, extensive API surface, database migrations, observability (OpenTelemetry), GitHub app integration. It’s actively developed — the last few commits were about refactoring agent briefing generation, splitting service lifecycle backends, and surfacing terminal provider failures. Real production shape.
Want me to dive into a specific package or subsystem?