Guides
A guide is a walkthrough: one outcome, shortest path, followed start to finish. A reference page lists every option. A guide picks the ones you need and gets you to a working result. When a guide and a reference disagree about a default, the reference is right — that is what it is for.
Get one reply working first with Install and Quick Start. Then pick the row that matches what you are trying to do.
The groups below are the same ones in the sidebar, and every guide in the tree appears here. The two cannot drift, because the docs build fails on a page that is not listed.
Connect a channel
Use Settings → Channels in the WebUI for guided setup. These guides cover the part the WebUI cannot do for you: the account, bot, token and permission steps on the platform's own side.
| Goal | Guide |
|---|---|
| Understand the shape all channels share | Chat app AI agent |
| Connect Telegram | Telegram AI agent |
| Connect Discord | Discord AI agent |
| Connect Slack | Slack AI agent |
| Connect Mattermost | Mattermost AI agent |
| Connect WhatsApp | WhatsApp AI agent |
| Connect Email | Email AI agent |
| Connect Signal | Signal AI agent |
| Connect Matrix | Matrix AI agent |
| Connect Microsoft Teams | Microsoft Teams AI agent |
Platform prerequisites and manual JSON for every channel: Channels.
Choose a model
| Goal | Guide |
|---|---|
| Point nanoinfra at any OpenAI-compatible endpoint | Configure an OpenAI-compatible provider |
| Keep answering when a provider fails | Configure model fallback |
| Stop Ollama re-reading the whole prompt each turn | Configure Ollama prompt caching |
Which provider to pick and why: Providers and Models and the Provider Cookbook.
Add a capability
| Goal | Guide |
|---|---|
| Give the agent tools from an MCP server | Configure MCP tools |
| Let the agent search and fetch the web | Configure web search |
| Give it memory that survives the session | AI agent memory |
| Have it work a goal over hours or days | Long-running AI agent |
Lock it down
Read these before exposing nanoinfra to anything but your own machine.
| Goal | Guide |
|---|---|
| Bound what local tools may read and run | Secure a local AI agent |
| Make an approval name a person, not a shared token | Named approvers with a proxy |
The policy behind both: Capability Gates and Identity and SSO.
Run it for real
| Goal | Guide |
|---|---|
| Work in the browser workbench | AI agent WebUI |
| Run it on your own hardware, unattended | Deploy nanoinfra gateway |
| Trace turns, tools and cost | Configure Langfuse observability |
Serving separate customers rather than separate bots: Multi-Tenancy.
Build on it
| Goal | Guide |
|---|---|
| Drive nanoinfra from Python | Python AI agent SDK |
Expose /v1/chat/completions to your own code | OpenAI-compatible agent API |
Another way through
The sidebar groups pages by subject and keeps these guides as one track. Tags cut across both, so you can ask the questions neither grouping answers:
- Every guide · every reference page · every concept page
- Everything about one platform: Slack, Telegram, MCP
- Everything on one subject: security, channels, operations