convez.ai
AgenteAutomaçãoAvançado
agent-team-foundation

First Tree

agent-team-foundation/first-tree

Agentes de IA conscientes do contexto do time, com loop humano controlado

Ver no GitHub
BomPluginScore · de 0 a 100

Framework open-source para agentes de IA que trabalham com contexto compartilhado do time. Integra-se ao GitHub e mantém uma árvore de contexto (decisões, responsabilidades, restrições) que alimenta o trabalho dos agentes e captura resultados duráveis.

Instalação

$npx convez-ai agent-team-foundation/first-tree

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
161
Forks
37
Acompanhando
1
Saúde
87%
Último commit
há 8 horas
Versão
v0.5.23
Issues abertas
327
Contribuidores
7

Casos de uso

  • Automatizar tarefas recorrentes mantendo contexto da equipe
  • Encaminhar trabalho para o agente correto com histórico de decisões
  • Criar ciclos de trabalho humano-IA com rastreabilidade

Veredito

First Tree resolve um problema real: agentes isolados carecem de contexto institucional. A ideia de Context Tree como memória compartilhada é sólida, mas a adoção exige disciplina do time para manter a árvore atualizada. Para equipes já versionando decisões e contexto, é um ganho direto.

README

A tradução para o português deste README ainda está na fila.

First-Tree

Context-grounded agentic work for teams.

First Tree is an open-source workspace where AI agents work from your team's shared context, not isolated prompts.

At the center is Context Tree: a team-maintained memory of decisions, ownership, repos, responsibilities, constraints, and prior work. Agents read it before they work; useful outcomes can flow back into it after the work is done.

The result is a human-agent work loop where every task can start with more team context, and every useful outcome can make the next task smarter.


Why First Tree

There are many AI agent workspaces. First Tree's core difference is the context loop:

user intent -> read team context -> context-aware agent work
-> human review/control -> durable outcome -> updated team context

This loop solves the part that usually breaks when teams start using agents:

  • context stays private in one person's terminal, prompt, or notes
  • agents repeat old decisions because they cannot see team memory
  • work finishes in a chat, PR, or document but never updates the shared context
  • humans get pulled in without enough background to make a good decision
  • each new agent task starts cold, even when the team already learned something

First Tree is built around a different assumption:

Agent work should be grounded in team context, visible while it runs, guided by humans at the right moments, and turned into durable outcomes that can update team memory.

That makes First Tree useful in two modes:

  • Focused copilot work - collaborate deeply with an agent in a persistent work stream for design, coding, writing, research, or problem solving.
  • Parallel review work - let agents move multiple tasks forward and involve you only when a decision, blocker, approval, or review is needed.

In both modes, First Tree gives teams one place to:

  • give agents the same Context Tree your team uses
  • start and continue agent work in persistent chats
  • see active work, blocked states, and human review points
  • inspect the process, artifacts, and rationale behind a request
  • turn useful outcomes back into updated team context

It is not another agent, and not just another workspace. It is a context loop for human-agent teams.

How It Works

First Tree connects five pieces around Context Tree:

  1. Context Tree — a Git-native team memory layer for decisions, ownership, responsibilities, constraints, and shared context.
  2. Web workspace — the daily surface for chats, agents, team members, computers, GitHub, and context-backed work.
  3. CLI + daemon — signs a computer in and keeps local agents connected.
  4. Agent runtime — runs agents on your machine and routes messages through First Tree.
  5. GitHub integration — connects code work, pull requests, and review back to the workspace.

Together, these pieces keep agent work connected to team context before, during, and after execution.

Get Started

Open the app at https://cloud.first-tree.ai (or your own deployment) and sign in. Learn more at https://first-tree.ai. The guided setup walks you through the first run: name your team, connect a computer, create your first agent, connect code, and start work.

See the Quickstart for the full walkthrough.

At the "connect a computer" step, setup gives you the command to install the CLI and link the machine:

npm install -g first-tree
first-tree login <connect-token>

The binary lives at first-tree; a short alias ft is also installed.

CLI

first-tree
├── login <token>           Sign this computer in
├── logout                  Stop the daemon and clear credentials
├── status                  CLI / daemon / server / auth overview
├── doctor                  Cross-subsystem readiness check
├── upgrade                 Upgrade to the latest published version
├── agent ...               Agent management
├── chat ...                Chats and messaging
├── org ...                 Organization-level operations
├── daemon ...              Background daemon lifecycle
├── config ...              View / modify this machine's client.yaml
└── tree ...                Context Tree onboarding, validation, automation

Run first-tree <namespace> --help for the full list under any namespace.

Repo Layout

  • apps/cli/ — published CLI (first-tree / ft)
  • packages/shared/ — Zod schemas, types, config system (@first-tree/shared)
  • packages/server/ — Fastify API server (@first-tree/server)
  • packages/client/ — Agent SDK + Runtime (@first-tree/client)
  • packages/web/ — React web workspace (@first-tree/web)
  • packages/e2e/ — black-box e2e harness (@first-tree/e2e)
  • skills/ — repo-local skill payloads for First Tree agents

Documentation

Development

pnpm install                                # Install dependencies
docker compose up -d                        # Dev PostgreSQL
pnpm --filter @first-tree/server dev        # Server
pnpm --filter @first-tree/web dev           # Web workspace
pnpm check && pnpm typecheck                # Lint + type check
pnpm test                                   # Tests
pnpm coverage                               # Local unit coverage
pnpm coverage:summary                       # Summarize generated coverage

See AGENTS.md for architecture, conventions, and the per-package development workflow. See CONTRIBUTING.md for the PR workflow.

License

Apache 2.0

Ver o original no GitHub

Bom para

Relacionados