ml-intern vs Terrarium

ml-intern is Hugging Face's own open-source autonomous ML engineering agent. Given a task description, it reads relevant papers, writes training code, executes it in a sandboxed environment, evaluates the results, iterates, and ultimately uploads a trained model to the Hugging Face Hub — with no human in the loop beyond the initial prompt. Under the hood, the agent runs an agentic loop of up to 300 iterations, using Claude as its reasoning backbone alongside smolagents. It has integrated access to HF documentation search, paper retrieval, GitHub code search, and sandboxed Python execution. When the context window fills (at 170k tokens), it auto-compacts rather than failing, and full sessions are uploaded to HF for inspection and reproducibility. What's notable here isn't just the capability — it's the source. Hugging Face is essentially shipping a proof-of-concept that the job of "write the ML training script, run it, fix it until it works, upload the result" can now be delegated to an agent. With 688 stars and active development as of this week, ml-intern is HF eating its own dog food on autonomous AI engineering. The "doom loop detector" that flags repetitive tool-use patterns is a candid acknowledgment of how agentic loops fail in practice.

Terrarium is a multi-turn evaluation and optimization engine for LLM agents built by evolvent-ai. Unlike static benchmark suites that measure agents against fixed input-output pairs, Terrarium creates persistent, stateful "living environments" — simulated deployment contexts where agents operate over extended sessions, accumulate state, use tools, and interact with simulated external systems. You evaluate agents the way you'd test a car: by driving it, not by measuring its doors. The system supports configurable environment complexity, including simulated databases, APIs, file systems, and user personas. Agents are scored not just on final outputs but on trajectory quality — how efficiently they reached the answer, how often they hallucinated intermediate steps, and how well they recovered from dead ends. The engine also supports continuous optimization loops where poor-performing trajectories trigger automatic prompt refinement. With 17 stars and created April 14, Terrarium is extremely new. But it's addressing a genuine gap: the disconnect between how agents perform on static benchmarks versus how they behave in production. As enterprise AI deployments scale, the need for realistic pre-production evaluation is becoming critical.