convez.ai
MCP ServerAutomaçãoAvançado
phuetz

Code Buddy

phuetz/code-buddy

Agente de IA para código que roda grátis na sua máquina com 15 provedores

Ver no GitHub
BomPluginScore · de 0 a 100

Agente de IA open-source multi-provider para desenvolvimento, com interface terminal TUI, aplicativo desktop Cowork e serviço autônomo 24/7. Roda localmente em Ollama (gratuito) ou integra 15 provedores de LLMs diferentes, oferecendo ~110 ferramentas, suporte a voz/visão e compatibilidade com MCP.

Instalação

$npx convez-ai phuetz/code-buddy

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
51
Forks
13
Acompanhando
2
Saúde
85%
Último commit
há 1 mês
Versão
v2.2.0
Issues abertas
37
Contribuidores
9

Casos de uso

  • Desenvolvimento de código com assistência local e autônoma
  • Automação de tarefas via serviço em background 24/7
  • Busca na web, análise de documentos e execução de shell commands

Veredito

Ferramenta robusta e ambiciosa para developers que buscam IA local com fallback para provedores em nuvem. A proposta de autonomia 24/7 free-first é diferenciada, mas requer setup técnico significativo e familiaridade com Ollama e TypeScript.

README

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

Code Buddy

Your AI-Powered Development Tool & Personal Assistant

The open-source AI coding agent that runs free, on your own machine.

Code, commands, web, voice, and vision — from your terminal, a desktop app, your phone, or a 24/7 autonomous service. Use local Ollama models at $0, or bring any of 15 providers. No lock-in.

  • 🆓 Free & local-first — runs entirely on local Ollama ($0), or any of 15 providers (Claude, GPT, Grok, Gemini, …) with auto-failover. Or log in with ChatGPT Plus/Pro for a flat-fee brain — no API metering.
  • 🧠 Reasoning you can watch — local models think step-by-step on screen, then act — see the live captures.
  • 🖥️ Runs everywhere — terminal TUI, the Cowork desktop app, an HTTP/WebSocket server, your phone, or a 24/7 background service — all on the same core engine.
  • 🤝 Multi-AI Fleet — multiple peers observe each other live and call each other's models & read-only tools (peer.chat / peer.tool.invoke) across your network.
  • 🤖 Autonomous — buddy autonomy install runs a self-driving service that claims & executes tasks free-first on local models, 24/7.
  • 🛠️ ~110 tools, skills & MCP — edit, shell, web search, browser, PDFs/Office, a skills marketplace, and MCP connectors to extend it.
  • 👁️ Personal companion — bidirectional voice, opt-in camera/presence (MediaPipe), persistent memory, and 20+ messaging channels.

Quick Start | Cowork + Companion | Features | FAQ | Documentation | Contributing


What is Code Buddy?

Code Buddy is an open-source multi-provider AI coding agent with a terminal UI, HTTP/WebSocket server, and Cowork desktop app. It supports 15 LLM providers with automatic failover and per-provider circuit breakers. It works as both a development tool (reads files, writes code, runs commands, creates PRs, plans complex tasks) and a personal companion (bidirectional voice conversation, durable memory, opt-in camera perception, screen/presence context, push notifications via 20+ messaging channels, and 24/7 background operation). With buddy login, a ChatGPT Plus / Pro subscription can become the flat-fee brain of the system without API-key metering.


In action

Free local AI, with the reasoning on screen. A local Ollama model (qwen3.6:35b-a3b) thinks through a task, then uses tools to do it — no cloud, ~$0.0001. Unedited captures from the Cowork desktop app:

More desktop demos (Fleet, Autonomy, Companion, …): cowork/readme.md.


ChatGPT Pro / Plus subscription login — buddy login, sign in once with your ChatGPT account, then chat with gpt-5.5 directly from the terminal. No API key, cost reported as $0.0000 (flat-fee plan).

Interactive TUI + tool calling. The agent reads project context, calls tools in parallel (web_search ×2 for the screenshot below), and streams the synthesised answer.

Self-audit. Asked to find a bug in its own integration code, gpt-5.5 reads src/codebuddy/providers/provider-chatgpt-responses.ts, identifies a stale-variable issue (mutated body.model not propagated), and proposes the exact fix:

More captures + walk-through: docs/screenshots/.


Quick Start

# Install — from source (recommended during the 1.0 release-candidate phase: gets the latest)
git clone https://github.com/phuetz/code-buddy.git
cd code-buddy && npm install && npm run build && npm link   # exposes `buddy` globally

# Or install the published release from npm
# ⚠️ during rc, the npm release can lag the source — prefer from-source for the newest features
npm install -g @phuetz/code-buddy

# Option A — bring your own API key
export GROK_API_KEY=your_api_key   # or GEMINI_API_KEY / OPENAI_API_KEY / ANTHROPIC_API_KEY
buddy

# Option B — log in with your ChatGPT Plus / Pro subscription (no API key needed)
buddy login                         # opens browser for OAuth → tokens persisted
buddy whoami                        # ✅ connected · your.email@example.com · Plan: pro
buddy                               # auto-routes to gpt-5.5 via the Codex backend, cost $0.0000

# Or with a specific task
buddy --prompt "analyze the codebase structure"

# Full autonomy
buddy --yolo

See Getting Started for installation options, headless mode, session management, and typical workflows. The ChatGPT Codex OAuth flow is documented with screenshots in docs/screenshots/. The desktop cockpit has a public overview in Cowork Desktop.


Cowork Desktop + Buddy Companion

Cowork is the desktop cockpit for Code Buddy: chat, tools, traces, workflows, settings, permissions, models, MCP connectors, skills, artifacts, and companion controls all run against the same core agent as the CLI. The Code Buddy settings panel can probe the local backend, start it when needed, discover models, and route Cowork turns through the embedded engine or a configured server endpoint.

# First-time identity and flat-fee brain route
buddy login
buddy companion setup

# Local backend for Cowork, Fleet, and OpenAI-compatible clients
buddy server --port 3000

# Launch the desktop app from an installed build
buddy gui
# or
buddy desktop

# Source checkout dev loop
npm install
npm run build
npm run dev:gui

Buddy companion commands are available in both the CLI and Cowork panel:

buddy companion status
buddy companion self
buddy companion evaluate
buddy companion radar
buddy companion impulses
buddy companion missions sync
buddy companion missions run-next
buddy companion safety recent
buddy companion camera status
buddy companion camera snapshot
buddy companion percepts recent

The camera bridge is explicit and local: snapshots are opt-in, percepts are append-only under .codebuddy/companion/, and Cowork uses MediaPipe Tasks Vision for face, hand, finger-tip, and pose signals. Face enrollment/presence recognition lives in Cowork's presence bridge and uses the local MediaPipe/Buffalo_S pipeline.

For low-latency voice experiments, Cowork can route STT/TTS through a local Kyutai DSM / moshi-server endpoint while keeping faster-whisper and Piper as fallbacks:

$env:COWORK_VOICE_PROVIDER='kyutai'
$env:COWORK_KYUTAI_URL='ws://127.0.0.1:8080'
npm run dev:gui

Use Buddy companion's Inspect voice action to probe the active route, Kyutai STT/TTS websocket reachability, ffmpeg availability, and the Piper/faster-whisper fallback state from the Electron app.

From source, Cowork requires Node.js >=22 in cowork/; the root CLI still supports Node.js >=18.


Features

CategoryHighlightsDocs
AI Providers15 providers (Grok, Claude, GPT, Gemini, Ollama, LM Studio, AWS Bedrock, Azure, Groq, Together, Fireworks, OpenRouter, vLLM, Copilot, Mistral), circuit breaker, model pairsproviders.md
Tools~110 tools with RAG selection, multi-strategy edit matching, Codex-style apply_patch, streaming, BM25 tool search, code exec sandboxtools-reference.md
Commands190+ slash commands, CLI subcommands (/goal, /dev, /pr, /lint, /switch, /think, /batch, /watch, /conflicts, /vulns, /replace)commands.md
Cowork DesktopElectron cockpit, embedded Code Buddy engine, backend health/start controls, model settings, permission rules, visual workflows, traces, artifacts, MCP/skills/plugin managementCowork Desktop, cowork/readme.md, cowork/ARCHITECTURE.md
Buddy CompanionChatGPT-backed identity, voice/TTS, proactive check-ins, self-evaluation, competitive radar, mission board, learned routines, safety ledger, local percept journalcommands.md
Vision & PresenceOpt-in webcam snapshots, MediaPipe face/hand/pose/finger-tip analysis, local face enrollment, presence state for the agentcowork/ARCHITECTURE.md
AgentsMulti-agent orchestration (5-tool API), 8 specialized agents, SWE agent, planning flow, A2A protocol, batch decomposition, agent teamsagents.md
ReasoningTree-of-Thought + MCTS (4 depth levels), extended thinking, auto-escalation, /think commandreasoning.md
SecurityGuardian Agent (AI risk scoring), OS/Docker/OpenShell sandbox, SSRF guard, secrets vault, write policy, exec policy, loop detection, omission detection, output sanitizersecurity.md
Channels20+ messaging channels (Telegram, Discord, Slack, WhatsApp, Signal, Teams, Matrix, IRC, and more), DM pairing, send policychannels.md
Context EngineSmart compression, tool output masking, image pruning, transcript repair, pre-compaction flush, restorable compression, JIT context, importance-weighted windowcontext-engine.md
Fleet & AutonomyPeer-to-peer hub (peer.chat / peer.tool.invoke / peer_delegate), A2A + ACP + MCP interop, 24/7 autonomous service (buddy autonomy install), event-driven daemon, claim TTL/lease, DAG task dependencies, workers→verifier→synthesizer swarm, free-first local→Tailscale→paid model tier, judge-gated goal loops (/goal Ralph loop, buddy goal, board goal-mode, peer-session goals)fleet-guide.md
InfrastructureHTTP server (OpenAI-compatible), WebSocket gateway, daemon mode, cron, device nodes, canvas/A2UI, 6 cloud deploy configs, MCP, pluginsinfrastructure.md
ConfigurationEnv vars, TOML config with profiles, model-aware limits, per-agent params, i18n (6 locales), personasconfiguration.md
DevelopmentTypeScript strict, Vitest (27,334 tests), ESM, middleware pipeline, facade architecturedevelopment.md

Additional Capabilities

  • Voice: 7 TTS providers, wake word detection, voice-to-code pipeline, hands-free companion conversation
  • Companion Loop: readiness checks, self-state snapshots, self-evaluation, competitive radar, impulses, missions, routines, and safety review
  • Vision: local camera snapshots, MediaPipe face/hand/pose/finger-tip percepts, face enrollment, and presence state for context-aware collaboration
  • Memory: Persistent + semantic + prospective + decision + coding style memory, ICM cross-session memory
  • Knowledge: Knowledge base injection, 40 bundled skills, self-authoring skills at runtime
  • Goal Loops (Ralph loop): /goal <text> + /subgoal — a judge model re-checks completion after every turn and auto-continues until done (turn budget, pause/resume, fail-open judge); also headless (buddy goal), on fleet board tasks (--goal-mode, blocks for human review when the budget is spent), and on peer sessions (peer.chat-session.goal)
  • Git Workflow: Auto-commit (Aider-style), /pr creation, merge conflict resolver, ghost snapshots
  • Code Intelligence: LSP rename/refactor, auto-import, bug finder (25+ patterns, 6 langs), OpenAPI generator, log analyzer
  • IDE Integration: VS Code extension (diff view, inline edit, model switch), JetBrains plugin, LSP server
  • Inline Context: @web, @git, @terminal mentions for contextual references
  • Workflows: Lobster typed DAG engine with approval gates, pause/resume tokens

Documentation

DocumentDescription
Getting StartedPrerequisites, install, first run, headless mode, session management
Application Validation GuideFull-app validation matrix with safe commands, evidence, and screenshot rules
ProvidersAll 15 providers, connection profiles, model pairs, circuit breaker
Tools ReferenceTool categories, RAG selection, edit matching, apply_patch, streaming
CommandsAll slash commands, CLI subcommands, companion commands, global flags
Cowork DesktopGitHub-visible overview, ChatGPT gpt-5.5 route, real validation commands, screenshot privacy policy
Autonomous Coding + Cowork ProgressEnd-to-end autonomous task execution, RunStore artifacts, and Cowork progress tracking
Cowork READMEDesktop installation, features, source build, sandbox modes
Cowork ArchitectureElectron contexts, bridges, embedded engine, persistence, runner model
AgentsMulti-agent orchestration, roles, SWE agent, planning flow, A2A
ReasoningExtended thinking, Tree-of-Thought, MCTS, /think command
SecurityPermission modes, Guardian Agent, sandboxing, SSRF, secrets vault
Channels20+ messaging channels, DM pairing, send policy
Context EngineCompression, tool output masking, JIT context, pre-compaction flush
InfrastructureHTTP server, WebSocket gateway, daemon, cron, deploy, plugins
DeploymentProduction deployment — systemd, Docker, Kubernetes, reverse proxy, env vars, upgrades
PdfCommander MCP IntegrationRegister the PdfCommander headless MCP server in mcp.json, verify with buddy mcp test, drive PDF ops from the agent
Fleet GuideMulti-AI hub — /fleet listen + /fleet send peer.chat, env-driven multi-provider auto-detect, Tailscale lab examples, autonomous fleet protocol
Hermes / OpenClaw ParityWhere Code Buddy stands vs Hermes Agent & OpenClaw — shipped capabilities, externally-gated features, and the one open code gap
ConfigurationEnvironment variables, TOML config, project settings, model limits
DevelopmentBuild, test, architecture, coding conventions, adding tools

Validation Snapshot

Latest local verification for the Cowork + Buddy companion loop (2026-05-24):

npm run typecheck
cd cowork && npm run typecheck
cd cowork && npm run build:e2e
cd cowork && npx vitest run tests/kyutai-bridge.test.ts tests/voice-bridge.test.ts tests/tts-bridge.test.ts
cd cowork && npx playwright test e2e/cowork-smoke.spec.ts --reporter=line --workers=1
cd cowork && npx playwright test e2e/companion-panel.spec.ts --reporter=line --workers=1
cd cowork && npx playwright test e2e/recent-features-smoke.spec.ts e2e/companion-panel.spec.ts --workers=1
cd cowork && npx playwright test e2e/codebuddy-settings.spec.ts e2e/recent-features-smoke.spec.ts e2e/companion-panel.spec.ts e2e/companion-live.spec.ts --reporter=line --workers=1
cd cowork && npx vitest run

Result: root and Cowork typechecks passed, Vite E2E build passed with existing chunk/dynamic-import warnings, Kyutai/voice bridge unit tests passed (31 tests), focused companion panel Playwright passed (1 test), recent-features IPC smoke passed (9 tests), Cowork smoke passed (29 tests), focused companion/settings/recent-feature Playwright suite passed (11 passed, 1 live hardware test skipped unless COWORK_LIVE_COMPANION=1), and Cowork Vitest passed (205 files, 1318 tests).


Contributing

git clone https://github.com/phuetz/code-buddy.git
cd code-buddy
npm install
npm run dev          # Development mode
npm test             # Run all tests
npm run validate     # Lint + typecheck + test (run before committing)

See Development for architecture details and coding conventions.


License

MIT License - see LICENSE for details.


Report Bug | Request Feature | Star on GitHub

Multi-AI: Grok | Claude | ChatGPT | Gemini | LM Studio | Ollama | AWS Bedrock | Azure | Groq | Together | Fireworks | OpenRouter | vLLM | Copilot | Mistral

Ver o original no GitHub

Bom para

Relacionados