Settings → AI → Models
Register the language model providers and specific models that AI agents can use. MisterShell supports ten providers: Anthropic, OpenAI, Ollama, Azure OpenAI, Google, Mistral, xAI (Grok), Cohere, OpenRouter, and AWS Bedrock (API Key).
Each row binds a display name (what users see when picking a model) to a concrete provider + model ID + endpoint. You can register the same underlying model multiple times under different names if you want separate configurations (for example, pointing at different endpoints). Generation parameters such as temperature are set per agent on the Agents tab, not per model.
Table columns
| Column | Notes |
|---|---|
| Name | Display name shown in agent configuration. |
| Provider | The upstream provider. |
| Model ID | The provider-specific model identifier (for example, claude-3-opus-20240229 or gpt-4-turbo). |
| Default | Green check for the workspace default model. |
| Actions | Edit / Delete. |
Common tasks
Register a new model
- Click Add Model at the top right.
- Fill the form:
- Name — short descriptive label.
- Provider — pick from the dropdown.
- Model ID — the provider’s identifier for the model. A hint below the field lists examples.
- API Key — credential used when calling the provider. It is encrypted before it is stored and is never returned by the API.
- Base URL — optional custom API endpoint. Needed for self-hosted providers (for example, an Ollama server); leave blank to use the provider’s default.
- Context Window — the model’s maximum context size in tokens.
- API Version — shown for Azure OpenAI only (for example,
2024-06-01). - Verify TLS certificate — on by default. The toggle is hidden for providers where it does not apply (AWS Bedrock, xAI). If the model endpoint presents a certificate signed by an internal certificate authority, load that authority into CA Certificates.
- Set as default model — tick this to make it the workspace default; the previous default is un-ticked automatically.
- Click Save.
Edit a model
- Click the blue edit icon.
- Update any field. Leave the API key untouched to keep the stored value, or click Remove key to run the model without one (an Undo appears until you save). Changing the provider discards the stored key — a key issued by one provider is meaningless for another — so enter a key for the new provider.
- Click Save.
Delete a model
- Click the red trash icon.
- Confirm.
Agents that referenced the model become non-functional (warning icon on the Agents tab) until assigned another model.
Change the default model
Edit any model and turn on Set as default model. The previous default is automatically un-ticked. The default is used by:
- Newly created agents that do not pick a specific model.
- Chat conversations when no agent is selected.
Providers
| Provider | Notes |
|---|---|
| Anthropic | Claude models. Requires an Anthropic API key. |
| OpenAI | GPT family. Requires an OpenAI API key. |
| Ollama | Self-hosted models served through Ollama. Set the Base URL to your server. |
| Azure OpenAI | Your Azure deployment. Requires the deployment URL, key, and API Version. |
| Gemini models. | |
| Mistral | Mistral models. |
| xAI (Grok) | Grok models. |
| Cohere | Cohere models. |
| OpenRouter | Aggregated access to many models through one key. |
| AWS Bedrock (API Key) | Bedrock-hosted models using an API key. |
Secret storage
Provider credentials are encrypted at rest with MisterShell’s database encryption key. The UI displays *** in place of a saved credential. Leaving that placeholder or the API key field unchanged keeps the existing credential; entering a new value replaces it.
Permissions
- Read:
app.ai.read. - Create / edit:
app.ai.write. - Delete:
app.ai.delete.