Chat with ARCANE
Use ARCANE's compact language models when a task needs open-ended generation rather than a bounded System 1 decision. Compare ARC 1 LM v2 with the original ARC 1 LM in the model selector.
ARC 1 SLM
Causal decoder
Chat with ARC 1, ARCANE's small language model
Start the local SLM API to talk to the real TensorFlow model.
How this works
The page calls a same-origin proxy at /api/slm-chat, which forwards the selected model ID to the Python SLM server. The server keeps the available ARC 1 checkpoints loaded and samples from the model you select. Health metadata shows whether each checkpoint is loaded, trained, and ready before the page enables generation.
These are compact next-token continuation models designed to demonstrate ARCANE's local generation path; they are not instruction-tuned assistants. The bundled v2 checkpoint is trained for architecture validation, so use a task-trained checkpoint when factual reliability is required.
Local: from arcane-docs-web run npm run dev:with-slm. This starts Next.js and python examples/serve_slm_api.py together. Or start the API on port 8001 and set SLM_API_URL=http://127.0.0.1:8001 in .env.local.
The server discovers Models/arc1_lm_v2.* and Models/arc1_lm.* automatically. Set SLM_DEFAULT_MODEL=arc1-lm-v1 to change the initial selection.