Anaconda Desktop
Connect Hull to the local model server you run in Anaconda Desktop, from the TUI or VS Code.
Hull can discover the text-generation model served by Anaconda Desktop and connect to its local OpenAI-compatible endpoint. Hull imports the server connection for you, so you don't need to copy an API key or configure a base URL manually.
Official documentation: Anaconda Desktop
Supported platforms
Anaconda Desktop and Hull must run on the same supported computer.
| Operating system | Supported installation |
|---|---|
| Windows | Windows 11, x86-64 |
| macOS | macOS 13 or later, Apple Silicon |
| Linux | Debian or Ubuntu, x86-64 or ARM64 |
Remote backends, remote-only model endpoints, and non-interactive or headless setup are not supported. Complete setup in the Hull TUI or VS Code on the computer running Anaconda Desktop.
Set up Anaconda Desktop
- Download and install Anaconda Desktop from the official product page. See Anaconda's installation guide for platform-specific steps.
- Open Anaconda Desktop and sign in with your Anaconda account or your organization's assigned credentials.
- Select Model Catalog and filter for a Text Generation model. Prefer a model tagged Tool Calling for full Hull agent functionality.
- Select a quantization that fits your computer, then click Download. Anaconda's model catalog guide explains model types, hardware requirements, and quantization choices.
- Select Model Servers, choose the downloaded model and file, and start its server. Enable tool calling when the selected model and server configuration support it.
Hull only discovers Desktop state and connects to an existing server. Downloading or deleting models and creating, switching, starting, stopping, or deleting servers must be done in Anaconda Desktop. See Anaconda's model server guide.
Connect Hull
- Run
/connectin the Hull TUI. - Select Anaconda Desktop.
- Follow the setup dialog. Hull can open Anaconda Desktop; after making changes there, return and choose Check again.
- When the model server is ready, choose Connect to import its connection and make the served model available in the model picker.
To refresh an existing connection, run /connect, select Anaconda Desktop, and choose connect / refresh now after changing the model, server address or port, or server API key in Desktop. Hull re-discovers the active server and replaces its stored model and connection information.
Tool calling
When Desktop reports that the model server supports tool calling, Hull allows you to connect without an additional warning. Tool calling lets the model use Hull's tools to inspect files, edit code, and run commands.
If tool support is unavailable or cannot be detected, Hull shows a warning and requires confirmation before connecting. You can still use the model for text generation, but normal coding-agent actions are limited and may fail. For the best experience, choose a Text Generation model tagged Tool Calling and enable the server's tool-call support when available.
How keys are handled
Anaconda Desktop uses two separate credentials:
- Desktop management key: allows local discovery of Desktop models and servers. Hull reads it from Desktop's configuration only when needed and never copies it into Hull storage.
- Inference server key: authenticates chat-completion requests to the running model server. Hull imports this key into its normal provider authentication storage together with the local endpoint and model details.
Hull never asks you to paste either key. If the inference server key or endpoint changes, use Refresh to import the current values.
Disconnect
In VS Code, open Settings > Providers and select Disconnect for Anaconda Desktop. In a terminal, run hull auth logout and select Anaconda Desktop. Disconnecting removes only Hull's stored provider authentication and connection metadata. It does not stop Anaconda Desktop, stop the model server, or delete the downloaded model.
Use Anaconda Desktop itself to stop or change the server.
Troubleshooting
- Desktop is not detected: Install it from the official product page, or use Open Anaconda Desktop if it is installed but not running.
- Sign-in is required: Open Desktop, complete sign-in, and leave the Hull setup dialog open so it can detect the change.
- No model is available: Download a model whose type is Text Generation, not an embedding-only model.
- No server is available: Start the downloaded model from Model Servers in Desktop.
- The server changed: Use Refresh in Hull to replace the stored model, endpoint, and inference key.
- The server is unhealthy: Check the server status and logs in Desktop, then restart it there before refreshing Hull.