NanoAgent
rizwan3d/NanoAgent
Agent de IA local para coding com controle humano e revisão explícita
Ver no GitHubAgent de IA local-first para revisão e desenvolvimento de código em desktop, terminal, VS Code, Visual Studio e CI. Funciona dentro de repositórios reais, mantendo o desenvolvedor no controle total das mudanças. Realiza tarefas como compreensão de codebase, planejamento de alterações, edição de arquivos e automação de code reviews.
Instalação
npx convez-ai rizwan3d/NanoAgent1 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
- Compreender e navegar codebases desconhecidas rapidamente
- Automatizar code reviews em pipelines CI/CD
- Planejar e executar refatorações seguras com aprovação explícita
Veredito

NanoAgent é uma solução pragmática para automação de tarefas de desenvolvimento em múltiplos ambientes (CLI, VS Code, Visual Studio). Sua abordagem local-first e requisito de aprovação humana para ações sensíveis o diferencia, embora exija integração com o fluxo existente do desenvolvedor.
README
A tradução para o português deste README ainda está na fila.
NanoAgent gives developers an AI teammate that can work inside a real repository while keeping the human in control. Ask it to understand a codebase, plan a change, edit files, run validation, review a diff, or automate a pull request review from the same toolchain you already use.
It is built for practical engineering work: local projects, real shells, version-controlled memory, reviewable changes, and explicit approval for sensitive actions.
What You Can Do
- Understand unfamiliar code faster with repository-aware search, summaries, and focused file inspection.
- Turn feature requests and bug reports into planned, editable, testable changes.
- Run code review with a findings-first workflow for local diffs, files, pull requests, and merge requests.
- Switch between hands-on implementation, read-only planning, and read-only review profiles.
- Keep team knowledge in reviewable
.nanoagent/memoryfiles instead of hidden agent notes. - Save repeatable slash prompts in
.nanoagent/commandsand run them from the terminal, desktop app, or editor. - Use the provider that fits your budget and policy, from API-key providers to subscription sign-in and local Ollama.
- Bring the same agent into desktop, terminal, VS Code, ACP-compatible editors, and CI review automation.
Product Surfaces
Desktop App
Use the desktop app for a visual workspace with chat, model controls, profile switching, budget controls, permission prompts, activity output, and undo/redo for tracked file edits.
Terminal
Use nanoai when you want a keyboard-first agent for interactive work, one-shot prompts, piped input, quick reviews, and automation-friendly output.
VS Code
Use the VS Code extension to chat with NanoAgent, send selections, review files, review Git diffs, and apply suggested changes from inside the editor.
Visual Studio
Use the Visual Studio extension to keep a NanoAgent tool window inside Visual Studio while driving the local NanoAgent CLI over ACP.
CI Review Automation
Use the included GitHub Actions, GitLab CI, and Bitbucket Pipelines examples to run NanoAgent against pull request and merge request diffs, then post review comments back to your platform.
Built For Control
NanoAgent is designed for useful automation without silent surprises.
- Profiles separate implementation, planning, review, exploration, and delegated work.
- Permission rules decide what can run automatically, what asks first, and what is denied.
- Sensitive actions can require approval, including file edits, shell commands, network access, MCP tools, memory writes, and elevated operations.
- Tracked file edits can be undone and redone.
- Secret-looking values are redacted before logs, memory, audit records, and displayed tool output.
- Your workspace stays local; prompts and selected context are sent only to the model provider you configure when needed for a request.
Benchmarks
NanoAgent includes task-based benchmarks in benchmarks/ so we can measure real coding-agent behavior, not just chat-style answers. The benchmark suites currently cover bug fixing, repository understanding, patch quality, security review, tool safety, and a regression bundle that reruns the same tasks over time.
Tasks are defined in benchmarks/tasks/ and grouped by benchmarks/manifest.json. Each task can score the agent with response checks, validation commands, and diff constraints against small local fixtures or this repository itself.
Run the full benchmark set locally with:
python benchmarks/scripts/run_benchmarks.py --all --system --skip-preflight
Run the regression-only suite with:
python benchmarks/scripts/run_benchmarks.py --suite regression --system --skip-preflight
Generated summaries are refreshed in benchmarks/results/latest.md and benchmarks/results/latest.json.
Telemetry
NanoAgent sends anonymous product analytics to PostHog using built-in US Cloud defaults in code. You can still override Application:Telemetry:* settings, and /disableanalytics writes Application.Telemetry.Enabled=false to .nanoagent/agent-profile.json for the current workspace.
Collected:
- NanoAgent version
- OS family
- app surface such as CLI, Desktop, VS Code, Visual Studio, JetBrains, GitHub Actions, GitLab CI, or Bitbucket Pipelines
- execution environment (
localorci) - CI provider when detected (
github_actions,gitlab_ci,bitbucket_pipelines, or generic CI) - feature names used
- success and failure counts
- token and duration buckets
- daily runs and usage-time buckets
Never collected:
- prompts
- source code
- file paths
- repository names or URLs
- API keys
- terminal output
Provider Choice
NanoAgent supports OpenAI, ChatGPT Plus/Pro sign-in, Anthropic Claude Pro/Max sign-in, GitHub Copilot sign-in, OpenRouter, Kilo Code, Cerebras, Groq, Anthropic, Google AI Studio, Ollama, LM Studio, Ollama Cloud, and OpenAI-compatible providers.
Get Started
Download the latest desktop build from GitHub Releases, or install the CLI:
Release assets also publish SHA256SUMS and GitHub artifact attestations so you can verify both checksums and build provenance.
Desktop App
| Platform | Architecture | Download |
|---|---|---|
| Windows | x64 | Setup .exe |
| Windows | x64 | Portable .zip |
| macOS | Apple Silicon / arm64 | Download .zip |
| macOS | Intel / x64 | Download .zip |
| Linux | x64 | Download .zip |
| Linux | arm64 | Download .zip |
CLI Install
NuGet / .NET tool:
dotnet tool install --global NanoAgent.CLI
Direct release installer:
macOS / Linux:
curl -fsSL https://raw.githubusercontent.com/rizwan3d/NanoAgent/master/scripts/install.sh | bash
Windows PowerShell:
irm https://raw.githubusercontent.com/rizwan3d/NanoAgent/master/scripts/install.ps1 | iex
Then start NanoAgent:
nanoai
The tag-based release workflows publish only NanoAgent to NuGet.org.
Documentation
The technical guide lives in docs/documentation.md. It covers installation details, first-run onboarding, desktop and terminal workflows, VS Code and Visual Studio setup, ACP integration, CI review automation, codebase indexing, providers, permissions, MCP, memory, hooks, troubleshooting, release automation, and source builds.
License
Apache License 2.0. See LICENSE.txt.