Compresr.ai NEW

Compresr is an open-source context compression library for LLM pipelines and agents.It offers two compression modes: coarse-grained chunk selection to retrieve relevant chunks for a query, and fine-grained token-level compression to reduce context at token granularity.

The library integrates with agent frameworks and LLM APIs to compress conversation history, tool outputs, long documents, and lists.Compression reduces context length to lower inference latency and token costs while helping preserve downstream task accuracy.

Typical use cases include long-document analysis (for example, SEC filings) and multi-turn agent workflows where context size is a bottleneck.compresr supports common LLMs and provides tooling for pipeline integration and gateway deployment.

Key Features

• Open-source context compression library for LLM pipelines and agents
• Coarse-grained chunk selection mode to retrieve relevant chunks for a query
• Fine-grained token-level compression for token-granularity context reduction
• Integrates with agent frameworks and LLM APIs to compress conversation history, tool outputs, long documents, and lists
• Supports common LLMs and includes tooling for pipeline integration and gateway deployment

Use Cases

• Compress multi-turn customer support and virtual assistant conversations using compresr's coarse chunk selection and token-level compression to shrink conversation history and tool outputs, reducing inference latency and token costs while preserving response accuracy
• Improve retrieval-augmented generation (RAG) and long-document QA by compressing and indexing lengthy documents with coarse-grained chunk retrieval and fine-grained token compression, enabling more context to fit into LLM prompts for cheaper, faster, and more relevant answers
• Optimize autonomous agents and LLM pipelines by compressing agent state, previous turns, and external tool outputs so multi-turn workflows stay within context windows, cut token usage and latency, and maintain task performance across complex chains of reasoning

Who is it for?

• Machine learning engineers
• Nlp engineers
• Mlops engineers
• Product teams
• Startup companies