How can I help you today?
Powered by AI · Runs locally · No ads · 100% private
Powered by AI · Runs locally · No ads · 100% private
Choose your AI backend. The Server URL below is auto-filled with the default for each runtime but you can override it.
Enter any server URL — local, LAN (e.g. http://192.168.1.50:11434), or a remote server.
For OpenRouter, use your API key. For external servers, ensure CORS is enabled: OLLAMA_ORIGINS=*.
For cloud providers (OpenRouter, Groq, OpenAI, Together AI), enter your API key here, or let the local server proxy use a key from .env. Server-loaded keys are never returned to the browser. Get a free OpenRouter key at openrouter.com/keys.
Lets the model search DuckDuckGo before answering when it needs to. Available on all OpenAI-compatible backends (OpenRouter, Groq, OpenAI, Together AI, Llama.cpp, etc.) — not available for local Ollama since it can't reach the internet. Adds a short instruction to the system prompt; no extra API cost.
Type a model name or select from available models.
Quickly switch the AI's personality. Selecting a preset updates the System Prompt and Temperature below.
Defines the assistant's personality and behaviour. Editing this manually will deselect any preset.
Controls randomness. Lower = focused & deterministic. Higher = creative & unpredictable. This is sent live with every message.
Max tokens kept in context. AI models support various sizes (up to 128K for Gemma 4).
Show words as they generate in real-time.
Scroll to bottom as response streams.
Subtle audio feedback on send/receive.
Set up your local device profile. No account, no internet — just your device.
Only stored locally on this device.
A unique ID for this device. You can export it as a file.
Click to copy · Unique to this device
Export saves a prefrontal_profile.json file. Import loads a profile from
another device.
Pair this web client with a local Prefrontal Agent. The agent dials out to this server — no ports to forward, no firewall changes. Commands run in the agent's Podman sandbox, scoped to its workspace.
In the agent terminal run prefrontal-agent pair <token>
(or pair <token> inside its REPL). Waiting for the agent…
Start with ! for a shell command: !npm test. Multi-line file
write: write src/app.js⏎file content here.