Give Navin a mission

"Find the cause of this bug, fix it, run the relevant tests and explain the changes."

"Research these competitors, compare their offers with sources and turn the findings into a report."

"Watch this project for actionable changes and tell me when something needs my attention."

Navin connects the model to your files, terminal, browser and tools. It can plan the work, delegate focused tasks to subagents, check results and continue while the goal and configured limits allow it.

The interesting part is what carries forward: project knowledge, decisions, reusable skills and the context of the work.

Get started

Navin Desktop (Windows / macOS / Linux)

Download Navin Desktop for Windows, macOS or Linux

CLI

Linux / macOS / WSL

curl https://navin.live/install -fsS | bash

Windows PowerShell

irm 'https://navin.live/install?win32=true' | iex

Open a terminal in your project, then launch:

cd your-project
navin-cli

1. Open Settings with Ctrl+G and configure a provider with your API key or a local model endpoint.
2. Select a model and a mode: Ask, Plan, Agent, Review, Security or Debug.
3. Start with: "Read this project and explain how it works. Then suggest one useful improvement."

The software is free to use under its open source license. External model APIs and connected services may charge for usage. Installation and troubleshooting.

Build from source (gateway + WebUI)

To run or change Navin on your machine:

git clone https://github.com/Navinspire-ia/navin.git
cd navin
make install
make start

make install sets up missing system packages where supported, the Python backend and the WebUI. make start launches the gateway and the WebUI at localhost:5173.

If you skip system setup (NAVIN_SKIP_SYSTEM=1 or sh scripts/install.sh --no-system), install Python 3.11+, Git, Make, Node.js 18+ and npm first.

On Linux/macOS, the native sandbox also needs rustup (make native).

CLI from the checkout: .venv/bin/navin-cli (Windows: .venv\Scripts\navin-cli). Full installation guide.

One workspace, many kinds of work

You want to...	Navin brings
Ship software	Repository exploration, code editing, terminal commands, Git, browser preview, tests and review.
Understand a codebase	Project Graph, Code Index, definitions, references and impact analysis.
Research and extract	Web search, browser tools, scraping, structured extraction and source-based reports.
Create deliverables	Documents, presentations, spreadsheets, architecture diagrams and media workflows.
Work on growth	Leads, enrichment, SEO, marketing research and campaign preparation.
Handle professional workflows	Tender analysis, career research, meeting transcription, notes and action items.
Delegate a larger task	Focused subagents that work in parallel and bring results back to the main agent.
Keep work moving	Goal loops, scheduled tasks and Heartbeat checks while the gateway is running.

Some workflows need extra dependencies, configured integrations or a compatible model. See the capability map for settings and module guides.

_{Recorded CLI session, shortened: inspect → fix → 6 tests pass. Still image and walkthrough}

What makes Navin worth exploring

Context that survives the chat. Project memory and the code graph help the agent recover decisions and find relevant files. Optional episodic memory adds recall of earlier work. Memory · Project Graph

Tools that do the work. Files, shell, Git, browser automation, APIs and MCP connect reasoning to execution. Skills package repeatable workflows; plugins and integrations extend the toolset. MCP setup · Skills

Your choice of models. Connect OpenAI, Anthropic, Google, Mistral, DeepSeek, Qwen and other providers, or local endpoints through Ollama, LM Studio and vLLM. Route different tasks to different models. Available capabilities depend on the selected provider and model. Configuration

Longer tasks with explicit controls. Loops keep a goal moving; Heartbeat checks for useful follow-up work. Approvals, resource budgets, checkpoints and stop controls govern execution. Automations · Security

Local execution keeps your workspace on your machine. When you choose a remote model or connected service, the relevant request data is sent to that service.

From a goal to verified work

Open the interactive workflow by downloading the HTML and opening it in a browser. The diagram shows how execution and persistent context connect.

An agent that can learn from experience

Navin includes experimental, opt-in learning layers:

Layer	What it explores
Self-Evolve / Auto-Skills	Turn repeated failures into candidate skills, evaluate them and promote or roll them back.
World model	Learn local predictions of tool outcomes from recorded trajectories.
Policy learning	Evaluate suggestions for the next action against held-out cases.

These layers are off by default. Evaluation gates control activation, and publishing a skill for every project requires a human action. They do not fine-tune the chat model.

The ambition is increasingly capable autonomous agents. Navin does not claim to be AGI.

Skills evolution · World model · Policy learning

Connect the tools you already use

• Model access: your own API keys or local inference.
• Extensions: MCP servers, skills, plugins and custom tools.
• Channels: Telegram, Slack, Discord, WhatsApp, Email, Teams and more through configured integrations.
• Surfaces: CLI, WebUI and downloadable desktop applications.

This GitHub repository contains the public agent runtime, CLI, WebUI and supporting tools. Desktop packaging, the navin.live service, website infrastructure and release publishing are maintained separately.

Build Navin with us

Try one real task. Tell us where the agent helped and where it got stuck.

• Report a reproducible bug or propose a feature.
• Add a provider, improve a skill, build an integration or contribute an evaluation case.
• Improve a translation or share a workflow others can reproduce.
• Read CONTRIBUTING.md and the CLA before submitting a pull request.

Created by Navinspire IA, maintained by @aymenghad and the Navin contributors.

License

AGPL-3.0, with a commercial licensing option from Navinspire IA. Contributions follow the Contributor License Agreement.