Use nanobot with Models Hub: Edit config.json

Configure nanobot's providers and agents.defaults in config.json to point at Models Hub, with install steps, recommended models, and error fixes.

nanobot (project: HKUDS/nanobot) is a lightweight, open-source personal AI agent framework with a web UI, Telegram, Discord, and Slack channels, plus tool calling and long-term memory. It configures model providers through a single config.json file; point a provider at Models Hub and nanobot's conversations and automations draw on your Models Hub balance. This page covers only the connection step — see the project docs for everything else.

Three things you need

A Models Hub API key, the endpoint https://modelsok.com/v1 (must end with /v1), and a model name.

Prerequisites

uv installed.
A Models Hub API key from the Tokens page, in a group that can access the models you want.

Recommended models

Use case	Model	Pricing	Live status
First choice for everyday chat, a balance of speed and quality	`claude-sonnet-4-6`	Price	Availability and latency
Complex reasoning and long context	`gpt-5.5`	Price	Availability and latency
Cheap and fast, good for high-frequency tasks	`gpt-5.4-mini`	Price	Availability and latency

Model names are authoritative on the pricing page.

Setup

Install nanobot:
```
uv tool install nanobot-ai
```
Windows users need to add .local/bin to their environment variables to invoke nanobot directly:
```
$env:PATH = "$env:USERPROFILE\.local\bin;$env:PATH"
```
Run nanobot --version to confirm it installed.
Initialize the configuration:
```
nanobot onboard
```
This creates a config file in your home directory: %USERPROFILE%\.nanobot\config.json on Windows, ~/.nanobot/config.json on Linux/macOS.
Edit config.json to point providers and agents.defaults at Models Hub:
```
{
  "agents": {
    "defaults": {
      "model": "claude-sonnet-4-6",
      "provider": "modelsok"
    }
  },
  "providers": {
    "modelsok": {
      "apiKey": "your-modelsok-api-key",
      "apiBase": "https://modelsok.com/v1"
    }
  }
}
```
apiBase must end with /v1; model must be a model ID actually exposed in the Models Hub console. Multiple providers can be configured side by side; agents.defaults.provider picks which one is used by default. Note: newer versions also support defining multiple models under modelPresets and referencing one by name via modelPreset — the direct fields shown here are the backward-compatible form; check your installed version if unsure.

Verify

Start an interactive conversation:

nanobot

Type any question, such as "Hello, tell me what day of the week it is today." A normal response means the configuration worked. Then check Usage logs in the Models Hub console; a request under that model name appears within seconds, confirming traffic went through Models Hub.

Common errors

Symptom	Cause	Fix
`nanobot` command not found	Windows users haven't added `.local/bin` to PATH	Run the PATH command above and reopen your terminal
401 Unauthorized	`apiKey` isn't filled in correctly, or the token is disabled	Re-copy from the Tokens page
Model not found	`model` doesn't match a model ID in the Models Hub console, or `apiBase` is missing `/v1`	Check the model name and URL against the pricing page
Configuration changes have no effect	Config changes require a restart to take effect	Save the file and re-run `nanobot`

FAQ

How is it billed? Per input and output token actually consumed per request, at the price on each model's pricing page.

How is this different from using the official API directly? Same request format; only apiBase changes to the Models Hub endpoint.

How do I switch models? Edit agents.defaults.model in config.json, save it, and restart nanobot.