Chroma Context-1: Training a Self-Editing Search Agent
📰 Hacker News (AI)
Chroma Context-1 is a 20B parameter agentic search model that achieves retrieval performance comparable to frontier-scale LLMs at a fraction of the cost and up to 10x faster inference speed
Action Steps
- Understand the limitations of traditional retrieval pipelines
- Learn about agentic search and its application in multi-hop retrieval
- Study the architecture and training of Chroma Context-1
- Experiment with using Chroma Context-1 as a subagent in conjunction with a frontier reasoning model
Who Needs to Know This
This research benefits AI engineers and ML researchers working on large language models and information retrieval systems, as it provides a more efficient and cost-effective solution for multi-hop retrieval
Key Insight
💡 Chroma Context-1 is designed to decompose queries into subqueries, iteratively search a corpus, and selectively edit its own context to free capacity for further exploration
Share This
🚀 Chroma Context-1: a 20B parameter agentic search model that achieves comparable performance to frontier-scale LLMs at a fraction of the cost! 🤖
Key Takeaways
Chroma Context-1 is a 20B parameter agentic search model that achieves retrieval performance comparable to frontier-scale LLMs at a fraction of the cost and up to 10x faster inference speed
Full Article
# Chroma Context-1: Training a Self-Editing Search Agent·|·Chroma
[](https://www.trychroma.com/home)[](https://github.com/chroma-core/chroma)
* Products
Products
[Sync](https://www.trychroma.com/products/sync)[Database](https://www.trychroma.com/products/chromadb)[Agent](https://www.trychroma.com/products/agent)
[Docs](https://docs.trychroma.com/)[Enterprise](https://www.trychroma.com/enterprise)[Pricing](https://www.trychroma.com/pricing)[Research](https://www.trychroma.com/research)
* Resources
Resources
[Changelog](https://www.trychroma.com/changelog)[Updates](https://www.trychroma.com/updates)[Community](https://discord.gg/MMeYNTmh3x)[GitHub](https://github.com/chroma-core/chroma)[Package Search](https://www.trychroma.com/package-search)
Log in Sign up
Chroma Technical Report
March 26, 2026
* * *
# Chroma Context-1: Training a Self-Editing Search Agent
* * *
[Hammad Bashir](https://x.com/hammadtime)Chroma
[Kelly Hong](https://x.com/kellyhongsn)Chroma
[Patrick Jiang](https://x.com/patpcj)UIUC
[Zhiyi Shi](https://zhiyiscs.github.io/)UIUC
Retrieval pipelines typically operate in a single pass, which poses a problem when the information required to answer a question is spread across multiple documents or requires intermediate reasoning to locate. In practice, many real-world queries require multi-hop retrieval, in which the output of one search informs the next. Recent work has shown that frontier LLMs perform this multi-hop search effectively through a process known as agentic search, simply defined as a loop of LLM calls with search tools. This mode of search often comes with significant cost and latency due to their use of frontier-scale LLMs.
We introduce Chroma Context-1, a 20B parameter agentic search model derived from gpt-oss-20B that achieves retrieval performance comparable to frontier-scale LLMs at a fraction of the cost and up to 10x faster inference speed. Context-1 is designed to be used as a subagent in conjunction with a frontier reasoning model. Given a query, it produces a ranked list of documents that are relevant to satisfying the query. The model is trained to decompose queries into subqueries, iteratively search a corpus, and selectively edit its own context to free capacity for further exploration.
Latency Cost
Average across all evaluations
## —Table of Contents
* [Introduction](https://www.trychroma.com/research/context-1#introduction)
* [Key Techniques](https://www.trychroma.com/research/context-1#key-techniques)
* [Related Work](https://www.trychroma.com/research/context-1#related-work)
* [Synthetic Task Generation](https://www.trychroma.com/research/context-1#synthetic-task-generation)
* [Agent Harness](https://www.trychroma.com/research/context-1#agent-harness)
* [Model Training](https://www.trychroma.com/research/context-1#model-training)
* [Model Behavior](https://www.trychroma.com/research/context-1#model-behavior)
* [Inference](https://www.trychroma.com/research/context-1#inference)
* [Results](https://www.trychroma.com/research/context-1#results)
* [Future Directions](https://www.trychroma.com/research/context-1#future-directions)
* [Conclusion](https://www.trychroma.com/res
[](https://www.trychroma.com/home)[](https://github.com/chroma-core/chroma)
* Products
Products
[Sync](https://www.trychroma.com/products/sync)[Database](https://www.trychroma.com/products/chromadb)[Agent](https://www.trychroma.com/products/agent)
[Docs](https://docs.trychroma.com/)[Enterprise](https://www.trychroma.com/enterprise)[Pricing](https://www.trychroma.com/pricing)[Research](https://www.trychroma.com/research)
* Resources
Resources
[Changelog](https://www.trychroma.com/changelog)[Updates](https://www.trychroma.com/updates)[Community](https://discord.gg/MMeYNTmh3x)[GitHub](https://github.com/chroma-core/chroma)[Package Search](https://www.trychroma.com/package-search)
Log in Sign up
Chroma Technical Report
March 26, 2026
* * *
# Chroma Context-1: Training a Self-Editing Search Agent
* * *
[Hammad Bashir](https://x.com/hammadtime)Chroma
[Kelly Hong](https://x.com/kellyhongsn)Chroma
[Patrick Jiang](https://x.com/patpcj)UIUC
[Zhiyi Shi](https://zhiyiscs.github.io/)UIUC
Retrieval pipelines typically operate in a single pass, which poses a problem when the information required to answer a question is spread across multiple documents or requires intermediate reasoning to locate. In practice, many real-world queries require multi-hop retrieval, in which the output of one search informs the next. Recent work has shown that frontier LLMs perform this multi-hop search effectively through a process known as agentic search, simply defined as a loop of LLM calls with search tools. This mode of search often comes with significant cost and latency due to their use of frontier-scale LLMs.
We introduce Chroma Context-1, a 20B parameter agentic search model derived from gpt-oss-20B that achieves retrieval performance comparable to frontier-scale LLMs at a fraction of the cost and up to 10x faster inference speed. Context-1 is designed to be used as a subagent in conjunction with a frontier reasoning model. Given a query, it produces a ranked list of documents that are relevant to satisfying the query. The model is trained to decompose queries into subqueries, iteratively search a corpus, and selectively edit its own context to free capacity for further exploration.
Latency Cost
Average across all evaluations
## —Table of Contents
* [Introduction](https://www.trychroma.com/research/context-1#introduction)
* [Key Techniques](https://www.trychroma.com/research/context-1#key-techniques)
* [Related Work](https://www.trychroma.com/research/context-1#related-work)
* [Synthetic Task Generation](https://www.trychroma.com/research/context-1#synthetic-task-generation)
* [Agent Harness](https://www.trychroma.com/research/context-1#agent-harness)
* [Model Training](https://www.trychroma.com/research/context-1#model-training)
* [Model Behavior](https://www.trychroma.com/research/context-1#model-behavior)
* [Inference](https://www.trychroma.com/research/context-1#inference)
* [Results](https://www.trychroma.com/research/context-1#results)
* [Future Directions](https://www.trychroma.com/research/context-1#future-directions)
* [Conclusion](https://www.trychroma.com/res
DeepCamp AI