Claw Compactor
Verifiedby OpenCompress
Compresses LLM context such as code, JSON, logs, and chat history through a reversible 14-stage pipeline with no LLM calls, via a Python API and CLI, for agent developers.
- Access
- APICLI
- Auth
- None
- Skills
- 4
Skills
Fusion Pipeline Compression
Routes code, JSON, logs, diffs, and search results through content-aware stages to cut token counts without LLM inference.
Reversible Rewind Retrieval
Stores originals behind hash markers so the LLM can call a tool to retrieve any compressed section.
Chat Message Compression
Compresses chat message lists with cross-message deduplication before the per-message pipeline.
Workspace Commands
Benchmarks, compresses, deduplicates, and estimates tokens across a workspace, with a watch mode for file changes.
Related Agents
Headroom
by Headroom Labs
Compresses tool outputs, logs, RAG chunks, files, and history locally before they reach the model, as a Python or…
Workweave Router
by Weave
Routes each prompt to the optimal model (Anthropic, OpenAI, Gemini) in under 50ms via an on-box embedder. Router 2.0…
Langfuse
by Langfuse
Open-source LLM observability and analytics. Trace, evaluate, and debug LLM apps with prompt versioning, score…
Prismer Cloud
by Prismer.AI
Gives AI agents a hosted runtime for shared error-to-fix evolution, persistent memory, compressed web context, and…