# Poco

> Poco makes AI coding agents cheaper by compressing their context - it cuts the token usage and cost of Claude Code, Codex, and Cursor while preserving meaning.

Supported clients: Claude Code, Codex, Cursor.

## Install

```
curl -sSL https://get.thepoco.io | sh
```

macOS only at launch (arm64 or x86_64). The binary is unsigned, so the installer clears the quarantine flag itself - no manual Gatekeeper steps needed.

## Docs

- [Platform](https://thepoco.io/platform): Poco compresses the context your AI coding agents send, then shows you the rest: every MCP server and sub-agent they load, what each one costs, the redaction audit trail, and spend by person and project.
- [Benchmarks](https://thepoco.io/benchmarks): How much does Poco compress, and at what quality? Token-savings and fidelity benchmarks for prompt compression.
- [Security](https://thepoco.io/security): What Poco sends off your machine and why - prompts, commands, documents, transcripts - plus the retention setting, the redaction controls, and the env vars that switch each feature off.

## Blog

- [A glossary of AI coding agent terms](https://thepoco.io/blog/ai-coding-agent-glossary): Plain definitions for the terms that actually determine your AI coding agent's cost - tokens, context windows, context compression, MCP servers, system prompts, and prompt caching - and how they relate to each other.
- [Where your context window actually goes](https://thepoco.io/blog/where-your-context-window-goes): A segment-by-segment look at what fills an AI coding agent's context window - system prompt, tool definitions, MCP servers, skills, and the conversation - and which parts you re-pay on every single turn.
- [How Poco compresses your prompts (without garbling them)](https://thepoco.io/blog/how-poco-compresses-prompts): A look at how Poco shrinks the prompts you send to Claude Code, Codex, and Cursor - extractive sentence selection, content-type awareness, and a quality gate that fails safe to your original.

## Security

Poco sends prompt text, shell commands and their output, documents, and session transcripts to our server, always over TLS. See https://thepoco.io/security for retention limits and per-feature kill switches.