Zap Coding Agent
zap-coding-agent/zap-coding-agent
Agente IA Rust para coding com injeção inteligente de contexto, sem bloat
Ver no GitHubAgente de IA para coding construído em Rust com injeção de contexto orientada a skills, sem bloat de prompt. Funciona como binário único sem dependências de runtime, oferecendo interface TUI interativa ou modo single-shot para automação.
Instalação
npx convez-ai zap-coding-agent/zap-coding-agent1 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
Casos de uso
- Automação de tarefas de desenvolvimento com IA
- Integração em pipelines CI/CD
- Assistência interativa em projetos grandes
- Geração e refatoração de código com contexto preciso
Veredito

Ferramenta madura e focada para quem quer IA em coding sem overhead. Arquitetura em Rust e single-binary tornam a distribuição e manutenção triviais; suporte multi-provider (Anthropic, OpenAI, Gemini, Groq) garante flexibilidade.
README
A tradução para o português deste README ainda está na fila.
⚡ Zap Coding Agent
An AI coding agent built in Rust — skill-first context injection, a hard security boundary, and a single binary with no runtime.
Website · Docs · Architecture
Install
macOS / Linux — one-liner
curl -fsSL https://raw.githubusercontent.com/zap-coding-agent/zap-coding-agent/main/install.sh | bash
The script detects your OS and architecture, downloads the latest release, installs to ~/.local/bin, and patches your shell config if needed. On macOS it also runs codesign --sign - automatically.
| Platform | Binary |
|---|---|
| macOS Apple Silicon (ARM64) | zap-macos-arm64.tar.gz |
| macOS Intel (x86_64) | zap-macos-x86_64.tar.gz |
| Linux x86_64 | zap-linux-x86_64.tar.gz |
Windows x86_64
Download zap-windows-x86_64.zip from the latest release, extract, and move zap.exe somewhere on your PATH:
Expand-Archive zap-windows-x86_64.zip .
Move-Item pkg\zap.exe "$env:USERPROFILE\.local\bin\zap.exe"
Build from source
Requires Rust 1.75+.
git clone https://github.com/zap-coding-agent/zap-coding-agent
cd zap-coding-agent
cargo build --release
cp target/release/zap ~/.local/bin/zap
Quickstart
zap # interactive TUI
zap --goal "add tests" # single-shot (non-interactive)
zap --auto --goal "..." # skip all permission prompts (CI)
zap --sdk # JSON-lines remote control over stdin/stdout
First run will prompt for an API key and model. Use /provider to switch later.
Supported Providers
| Provider | Model examples | Auth |
|---|---|---|
| Anthropic | claude-sonnet-4-6, claude-opus-4-6 | API key |
| OpenAI | gpt-4o, gpt-4-turbo | API key |
| Google Gemini | gemini-2.0-flash, gemini-2.5-pro | API key or gcloud ADC (keyless) |
| LM Studio | gemma-4-e4b-it | None (local) |
| Groq | llama-3.3-70b-versatile | API key |
| OpenRouter | (various) | API key |
Configuration
All settings live in ~/.agent.toml. Environment variables take precedence.
Use /provider inside zap to switch interactively — settings are saved per provider.
# ~/.agent.toml — managed by zap /provider
provider = "anthropic" # active provider slug
permission_mode = "ask" # ask | auto | deny
# Optional: import skills from other tools
skill_paths = [".kiro/skills", ".claude/skills"]
# Optional: always-on context from project docs
steering_dirs = ["docs/decisions", ".wiki"]
[providers.anthropic]
kind = "anthropic"
model = "claude-sonnet-4-6"
api_key = "sk-ant-..."
[providers.openai]
kind = "openai"
model = "gpt-4o"
api_key = "sk-..."
[providers.lm_studio]
kind = "openai"
model = "gemma-4-e4b-it"
base_url = "http://localhost:1234/v1/chat/completions"
Environment variable overrides
AGENT_PROVIDER=anthropic AGENT_API_KEY=sk-ant-... AGENT_MODEL=claude-sonnet-4-6 zap
ANTHROPIC_API_KEY and OPENAI_API_KEY are also read automatically.
Google Gemini — keyless via gcloud
gcloud auth login
gcloud auth application-default login
zap
# /provider → select "Google Gemini"
# Auto-detected credentials show a "✓ ready" badge.
Or configure manually:
provider = "gemini"
[providers.gemini]
kind = "openai"
model = "gemini-2.0-flash"
base_url = "https://generativelanguage.googleapis.com/v1beta/openai/chat/completions"
credential_method = "gcloud_adc"
Set GOOGLE_API_KEY in your environment to use an API key instead.
Slash Commands
| Command | Description |
|---|---|
/help | List all commands |
/provider | Switch LLM provider and model |
/branch | Create a new conversation branch |
/compact | Summarize old turns to free context |
/skill | List, show, use, or unuse skills |
/init | Analyze the project and save context to ZAP.md |
/memory | View/edit session memory |
/tasks | Show the current task list |
/paste | Paste clipboard content |
/clear | Clear the screen |
/exit | End the session |
SLM Support
Zap is the only coding agent that runs local small language models as first-class executors for the mechanical bulk of coding work — while frontier models handle the thinking.
- Run local models via LM Studio, Ollama, or any OpenAI-compatible endpoint
- Structured plan execution — a frontier model designs the plan, your local model executes it step-by-step (100% success rate on pre-written plans in research)
- Pre-indexing —
zap --index-onlybuilds an AST index so SLMs navigate code withcode_map/find_definitioninstead of slow manual reads - Streaming watchdog — tolerates long local-model prefills (minutes) with progress notices and idle detection
- Core tool profile —
AGENT_TOOL_PROFILE=coresends only 6 tool schemas, making tool-calling tractable for small models
Full SLM support docs → · Research results →
Security
Zap has a hard security boundary:
- Permission modes:
ask(default),auto, ordeny— controls all tool access - Secret scanner: 25+ patterns (API keys, tokens, passwords) detected and redacted before sending to cloud
- Shell sandbox:
workdirmode confines shell to the project root;containermode wraps commands via Docker/Podman with--network none - Audit trail: Every tool call, permission decision, and LLM request is logged
See docs/SECURITY.md for the full threat model.
Architecture
See ARCHITECTURE.md for a module map and design decisions derived from the source code.
License
MIT