Waitlist — hosted API, private models, eval datasets

[GregoryBartolo]

Waitlist — hosted API, private models, eval datasets

If you build agents and want more than the public demo at dalek-ai-router-api.hf.space, this thread is the signup. Just reply with your handle and a one-liner on what you're building. No form, no email collection.

What's on the table

• Hosted API access beyond the shared CPU free tier: dedicated endpoints, custom rate limits, latency SLAs.
• Private model fine-tuning on your tool catalog and traces. The current default (baseline-v1-desc-hybrid-multilingual-next-v1) ships +20.6pp top-3 over plain retrieval, +27.7pp tau-bench LOSO. A model trained on your specific traces will go further.
• Eval datasets: the 14K trace corpus is public; the FR eval (50 queries) is up at dalek-ai/agent-tool-router-eval-fr. Custom evals (per-vertical, per-stack) can be built on request.
• MCP integration: native loaders for MCP server tool catalogs are on the roadmap.

How to register

Reply below with:

• your handle / GitHub
• 1 sentence: what agent / stack you're building (LangChain, custom, MCP, n8n, etc.)
• 1 line: what would make you switch from prompt-stuffing tool descriptions to a router call

That's it. I read every reply. If you want a quick async DM instead, ping @dalekkskaro on X.

Status

• Public SDK: agent-tool-router (MIT, GitHub).
• Hosted API: live at https://dalek-ai-router-api.hf.space — try it without signup.
• Models: 6 pretrained on https://huggingface.co/dalek-ai.
• Dataset: 14K agent traces + 50-query FR eval, both public.

— Dalek