AI Models

Configure the language models that power ClearFox.

1
Go to console.anthropic.com/settings/keys.
2
Click 'Create Key', give it a name (e.g. 'ClearFox'), and copy the key.
3
Paste the key in the ClearFox setup wizard (or Admin → LLMs after installation).
4
Recommended models: claude-sonnet-5 (balanced daily work), claude-opus-5 (default, complex reasoning).
5
⚠ Data retention — Fable 5: the provider requires 30-day data retention for this model (zero data retention unavailable). Use caution with sensitive data. ClearFox marks affected models with a ⚠ in the model picker.
6
Pricing: pay-per-token via Anthropic. No minimum commitment.

1
Go to platform.openai.com/api-keys.
2
Click 'Create new secret key', name it (e.g. 'ClearFox'), and copy the key.
3
Paste the key in the ClearFox setup wizard (or Admin → LLMs after installation).
4
Recommended models: gpt-5.6-luna (cheap, high volume), gpt-5.6-sol (flagship, 1M context).
5
⚠ Data retention — GPT-5.5 / GPT-5.5 Pro: OpenAI reserves the right to make these models ineligible for zero data retention and to keep prompts/outputs its classifiers flag as potentially policy-violating. Use caution with sensitive data. ClearFox marks affected models with a ⚠ in the model picker.
6
Pricing: pay-per-token via OpenAI. No minimum commitment.

1
Install Ollama: ollama.com/download (macOS, Linux, Windows).
2
Pull a model:
```
ollama pull llama3.1:70b
```
3
Ollama serves an OpenAI-compatible API on port 11434 by default.
4
If Ollama runs on the host machine, use host.docker.internal as the host.
5
Enter the Ollama base URL and model name in the ClearFox setup wizard (or Admin → LLMs).
```
Example: http://host.docker.internal:11434 with model llama3.1:70b
```
6
All data stays on your machine. No internet access required after model download.

Start vLLM with an OpenAI-compatible server:

vllm serve meta-llama/Llama-3.1-70B-Instruct --port 8000

Enter the vLLM base URL and model name in the ClearFox setup wizard (or Admin → LLMs).

Example: http://host.docker.internal:8000/v1 with model meta-llama/Llama-3.1-70B-Instruct

5
Requires a GPU with sufficient VRAM (70B model needs ~40GB). For smaller setups, use 8B or 13B models.
6
Full air-gap support. Nothing leaves your network.

1
Serve a GGUF model with llama.cpp’s llama-server, which exposes an OpenAI-compatible API (port 8080 by default):
```
llama-server -m /models/your-model.gguf --port 8080 --ctx-size 32768
```
2
In Admin → LLMs → llama.cpp (also available in the setup wizard), fill in Base URL. Enter the server root — ClearFox appends /v1 itself, so do not include it. If llama-server runs on the host machine instead of inside the Docker network, use host.docker.internal:
```
http://host.docker.internal:8080
```
3
API Key is optional — leave it empty unless you started llama-server behind an API key. When set, ClearFox sends it as a bearer token.
4
Models are auto-discovered. Once the base URL is filled in, ClearFox reads the server’s model list and shows it under Available Models. Tick the models you want available in chat and press Save. Use Refresh after you load a different model.
5
Size the context window at launch. ClearFox reads the context actually allocated by the server, and llama.cpp cannot grow it per request — an over-long prompt is rejected outright. Start llama-server with a --ctx-size (-c) value large enough for your longest prompts.
6
Everything runs on your own hardware. Nothing leaves your network.

1
Go to aistudio.google.com/apikey and create an API key.
2
Paste it into Admin → LLMs → Gemini → API Key and press Save. Gemini keys start with AIza.
3
There is no base URL to configure — ClearFox talks to Google’s OpenAI-compatible endpoint. Models are not discovered from the API: the Gemini models ClearFox supports appear in the model picker as soon as a key is saved.
4
Agents with a structured JSON output schema fall back to prompt-only JSON on Gemini. Google’s OpenAI-compatible layer rejects a request that combines tools with a strict response schema, and ClearFox always sends MCP tools. If an agent must guarantee schema conformance, run it on another provider.

1
Create an API key in the xAI console.
2
Paste it into Admin → LLMs → Grok → API Key and press Save. Grok keys start with xai-.
3
There is no base URL to configure — ClearFox uses the hosted xAI API. Models are not discovered from the API: the Grok models ClearFox supports appear in the model picker as soon as a key is saved.
4
Grok supports strict structured JSON output, so agents with an output schema get schema-enforced responses natively.