omni
View on GitHubYour agent pays twice for output it has already seen. OMNI returns a handle instead: 97.2% off a file read twice. Nothing deleted, nothing invented.
Rust CLI plus MCP server that hooks around an AI agent's tool calls to strip noisy output and replace already-seen bytes with retrievable handles. Every removed byte is archived in local SQLite and can be retrieved verbatim; original Apache-2.0 source included.
Use Cases
Cut token spend for coding agents by collapsing repeated file readsDistill noisy build/test/docker logs before the model sees themCompress git log/diff output while keeping every commitPersist project memory across sessions and editorsRetrieve byte-exact original output via handle after compressionRun as pre/post hooks around agent tool callsExpose omni_run and omni_retrieve over MCP to non-hook hostsReduce LLM context length without deleting or fabricating data
Built With
- Language
- Rust
- Frameworks
- MCP (rmcp) · Claude Code · Codex CLI · Gemini CLI · Cursor · Windsurf · Cline · Roo · OpenCode · VS Code · Zed · Copilot · Antigravity · Aider
Tags
context-compression · token-optimization · cost-reduction · cli · rust · mcp · hooks · claude-code · cursor · memory · prompt-caching · context-engineering · developer-tools · agentic-ai