Hugging Face opens 1M Spaces as callable tools for AI agents

What the tweet revealed

Clement Delangue wrote that Hugging Face is becoming a platform for agents and said they can “call 1M HF spaces”. The number is the signal: Hugging Face is trying to turn its hosted app catalog into an invocation layer for coding agents, research agents, and product builders that need executable tools.

Delangue is Hugging Face’s co-founder and CEO, so his account is usually a high-signal place for platform direction rather than a random showcase. The matching Hugging Face documentation describes Spaces as Agent Tools. Every compatible Gradio Space exposes a plain-text agents.md file that agents such as Claude Code, Codex, OpenCode, and Pi can read directly. The response gives a schema URL, a call template, a poll template, and an authentication hint, so an agent can discover how to call a Space without a hand-written integration.

Why the catalog matters

A million callable Spaces changes the agent discussion from model intelligence alone to tool liquidity. Instead of wiring an assistant to a small fixed list of APIs, an agent can search for a task-specific Space, inspect the interface, call the endpoint, and pass the result to another Space. Hugging Face’s documentation gives a concrete chain: generate an image with one Space, then pass that image into a 3D-model Space.

The practical constraint is trust. A large tool catalog raises questions about permissions, quota, provenance, cold-start latency, and whether agents can tell a maintained Space from an abandoned demo. The docs also point to HF_TOKEN for authenticated calls, which means billing and rate limits will become part of the agent runtime design.

What to watch next is whether Hugging Face adds ranking, safety metadata, and enterprise controls around these agent-callable Spaces. If those layers mature, Spaces could become a default tool registry for agents rather than only a place to try demos. Source: Clement Delangue source tweet · Hugging Face Spaces agent tools docs