Chroma Context-1: Training a Self-Editing Search Agent (80 minute read)

Chroma Context-1 is a 20B parameter agentic search model trained on over eight thousand synthetically generated tasks. It achieves retrieval performance comparable to frontier models at a fraction of the cost and with up to 10 times the inference speed. The model returns a ranked set of ing documents to a downstream answering model, cleanly separating search from generation. It decomposes high-level queries into sub-queries and iteratively searches a corpus across multiple turns, selectively discarding irrelevant results to free capacity as its context window fills.