Providers and models
Every way to connect a model: API keys, the plans you already have, and models running on your own machine.
Connect models in Settings → Models & API keys. Each host has its own providers and keys, so pick a server there to set up that server.
API keys
Paste a key and InkBot checks it and lists the models. Keys are stored in Windows Credential Manager on your PC, or encrypted on a server.
| Provider | Models |
|---|---|
| Claude (API) | Claude, with an Anthropic API key |
| ChatGPT (API) | GPT models, with an OpenAI API key |
| Gemini (API) | Gemini, with a Google AI Studio key |
| Grok (API) | Grok, with an xAI key |
| Kimi (API) | Kimi, with a Moonshot AI key |
| MiniMax (API) and MiniMax (plan) | MiniMax, pay-as-you-go or with an M Plan subscription key |
| OpenRouter | Hundreds of models behind one key |
| Nous Research | Models on your Nous Portal credits |
Your plans, through their own apps
These use the plan you already pay for. InkBot runs the vendor's own agent app, which signs in by itself, and shows its steps and approvals in the chat like any other bot.
| Provider | Through |
|---|---|
| Claude | Claude Code (claude auth login) |
| ChatGPT | Your ChatGPT plan, signed in from InkBot |
| Codex | OpenAI's Codex agent |
| Grok | Grok Build |
| Cursor | Cursor's CLI agent |
| Kimi | Kimi Code |
| OpenCode | The open-source coding agent, with any model, including free ones built in |
Install the app on the host first. InkBot never reads these apps' credential files. For OpenCode, the Accounts section lists its provider accounts, adds one (in OpenCode's own login window) and removes one.
Models on your own machine
- Local models: InkBot downloads open models from Hugging Face and runs them itself. See Local models.
- Ollama: point InkBot at your Ollama server.
- Unsloth Studio: its address and optional key. InkBot reads the context the model is loaded with.
Switching models on
After connecting, switch on the models you want to offer. Only switched-on models appear in the model picker and the bot editor.
Effort and thinking
Every bot has an effort level: Quick, Balanced, Deep or Max. For models that reason, InkBot maps it to the provider's own setting (Claude's adaptive thinking and effort, OpenAI's reasoning effort and so on). Quick uses the least of your plan or credits; Max thinks as much as the model allows.
Usage
Settings → Usage shows tokens and cost by day, bot and model for each host. Costs are at API prices where the provider reports them. Replies whose provider doesn't report tokens are estimated, and marked as such.