Core concepts
Threads
Conversations hold the user request, assistant response, selected model, files, and tool activity.
Tools
Built-in tools and MCP tools let the assistant search, execute code, query data, automate browsers, and call external services.
Agents
Agents package instructions, model preferences, and tool access into reusable assistants for focused tasks.
Workflows
Visual workflows connect LLM and tool nodes, then become callable tools inside chat.
How it fits together
What to read first
1
Choose how you want to run it
Use the hosted app, run locally with
pnpm, or deploy with Docker/Vercel.2
Add one model provider
Navigator only needs one working AI provider key to begin. Add more providers later for model choice and fallback.
3
Connect useful tools
Start with web search or one MCP server, then add presets and agents once the first flow works.
4
Create repeatable workflows
When a task becomes routine, move it into a workflow and invoke it with
@.Install Navigator
Set up dependencies, environment variables, PostgreSQL, and the local development server.