Lattice on the web: chat and IDE
New chat
workingWhat should we work on?
The model reads the files you add with @, proposes changes you review before anything is written, and runs no commands here.
Recent chats
Choose a model: one on this computer, or one with your own key.
Enter to send · Shift+Enter for a new line
Models make mistakes. Check what matters before acting on it. No request was sent until you press Enter, and none is retried on its own.
Your files, beside the conversation
Open a folder on this computer (Chrome or Edge) or add files. Edit them here, attach them to a message with @, and review what the model proposes before anything is kept.
What exists today, and where your words go.
What each kind of model does with what you send.
This computer (llama.cpp)
Start llama.cpp's server with your GGUF model, then Alelyon's loopback bridge (python www/tools/local_model_bridge.py from the source, or the app's own runtime). The bridge answers only Alelyon's pages and binds to this computer only. Prompts stay here.
Open models, your key
OpenRouter, Hugging Face, Together AI, Groq, Fireworks AI, DeepInfra and Cerebras, called straight from this tab with your key. What you send goes to that provider under its terms and prices; Alelyon never sees it or the key. Before the first message to each provider, this tab asks you.
Your files
A folder you open is read by this tab only. A file reaches a model only when you add it to a message, and a model's change reaches your folder only when you keep it. Secret-looking files (.env, keys) are listed but never opened.
The web chat cannot run commands, browse, use MCP servers or act on your desktop. Those belong to Lattice in the Alelyon app, where each asks you first.