Chroma Context-1: Training a Self-Editing Search Agent
Summary
Chroma's Context-1 is a 20B gpt-oss search subagent that returns ranked supporting documents instead of answers.
Separate search from generation.
Context-1 returns a ranked document set to a downstream reasoner. Missing evidence is worse than extra noise, because the generator can filter but cannot invent a document it never received.
missing a critical document is often worse than including an irrelevant one
We deliberately avoided lossy compression in this work, opting for document-level retention to preserve evidence fidelity.
Teach the agent to edit its own context.
A prune tool plus token-budget pressure lets the model discard irrelevant chunks mid-search instead of summarizing them, reducing context rot without lossy compression.
Train recall first, then precision.
Supervised warmup on large-model traces is followed by RL with a recall-heavy F1 that later anneals toward precision, plus trajectory-recall credit for documents later pruned.
purpose-trained search subagents represent a practical path toward making agentic search both more capable and more accessible
Verify synthetic tasks by extraction.
The judge extracts document and clue quote pairs and checks they appear verbatim, reaching more than 80% human alignment without full-document labeling.
Small specialized subagents generalize.
Trained only on web, legal, and finance, Context-1 still improves on held-out email and public benchmarks. Four parallel rollouts stay cheaper than one frontier call.
Context-1 operates as a retrieval subagent: rather than answering questions directly
Read the original ↗Save to your Sponge
Summarized with Sponge. Quotes are the author’s own words.