Load Module from URL

Ollama Endpoint: Connect

Current Model: Select a model Set Model

Agent Loop Optimizations: Enabled

⚡ Tool def caching, trivial query skip, parallel safe tools, debounced reflection
Disable for debugging or if model behaves unexpectedly. Reload after toggle.

Stream Tokens: Enabled

Show AI responses token-by-token. Disable for instant full responses and reduced crash risk.
Disable on slow connections or when browser becomes unresponsive during long outputs.

A separate lightweight AI that validates the main agent's tool results. Has its own endpoint and model — independent from your active AI model.
⚡ Best validator model: qwen2.5:0.5b (1.94s, 352MB)

Enable Validator: Disabled

Validator Endpoint: 🔌 Connect & Load Models

Validator Model: Connect to load models Save Validator Config

⚠️ Connection Notes
• Cloud models (deepseek-v4*, gemma4:*b-cloud) via Ollama proxy often ignore tool calls and answer from training data instead. Use local models for tool calling.
• To connect to local Ollama from a web page: set OLLAMA_ORIGINS on the server: export OLLAMA_ORIGINS="https://webx11.duckdns.org" then restart Ollama.
• Test local connection: open a terminal and run curl http://localhost:11434/api/tags
• If you see 403 errors in console: the OLLAMA_ORIGINS env var isn't matching your page URL.

Installed Skills

Create New Skill

Create Skill

Add / Update Config

🖥️

🎮 Start Tutorial Skip — I'll explore