How to Build a Personal AI Agent with Navin

This guide builds a personal AI agent you run locally in the Navin desktop app, then optionally connect to chat apps, memory, tools, and automations.

What you will build

Navin installed as a desktop app
one working model provider
one successful chat reply
a project you can keep using

When to use this

Use this when you want a personal AI agent that you control rather than a hosted chat-only interface. Navin is useful when the agent needs local project access, tool calls, session history, memory, scheduled work, or chat app delivery.

Install

Open navin.live/download.
Download the installer for Windows, macOS, or Linux.
Install and open Navin.
Complete the setup wizard (language, model access, first action).

If terminals and config files are new to you, also read ../start-without-technical-background.md.

First working chat

Open Settings → Providers and add an API key, or finish local Ollama setup under Settings → Providers → Ollama.
Open Settings → Models, add a configuration, and set it Active.
Open a project (demo workspace or your own folder).
Send a short message in chat, for example:

Hello! Summarize what you can help me with in this project.

Next steps in the app

Memory - keep durable facts with Dream (/dream in chat). See ai-agent-memory.md.
Channels - Telegram, Discord, Slack, and more under Settings → Channels. See chat-app-ai-agent.md.
Automations - ask for schedules in chat; leave Navin open. See ../automations.md.
MCP / Apps - external tools. See configure-mcp-tools.md.
Models - named presets, task routing, fallbacks under Settings → Models. See ../configuration.md.

Production notes

Keep one project per personal or client context.
Use model configurations when you want stable names for fast, deep, local, or fallback models.
Leave Navin running for chat apps and scheduled automations.
Prefer Settings over hand-editing files for everyday changes.

Security notes

Do not paste API keys into shared chats or public docs; use Settings fields.
Prefer chat app pairing for first setup. Keep allowlists narrow.
Enable workspace restriction before exposing file or shell tools to other users.
Use a separate project for experiments that can modify files.

Troubleshooting

No reply: open Settings → Providers and Settings → Models, confirm a configuration is Active, then retry chat.
Wrong files: check the project picker in the workbench.
Channel silent: keep Navin open and review Settings → Channels.