claudioemmanuel/squeez — Hook-based token compressor for 5 AI CLI hosts (Claude Code, Copilot CLI, OpenCode, Gemini CLI, Code…

Description and use

**Brief Summary**
Claudioemmanuel/squeez is a Rust‑based, zero‑dependency hook system that compresses token streams for five AI‑CLI hosts (Claude Code, Copilot CLI, OpenCode, Gemini CLI, Codex CLI), achieving up to 95 % reduction in Bash payloads. It adds a “signature‑mode” for safe code reads, cross‑call deduplication, an MCP (Model Context Protocol) server, and a self‑teaching protocol that lets AI agents learn from prior interactions.

**Value**

• **Token efficiency:** By shrinking prompts and responses, squeez cuts API costs and speeds up LLM calls, especially for long Bash or code snippets.
• **Standardised integration:** The MCP server and hook API give every supported AI CLI a common way to expose tools, data, and execution results, turning ad‑hoc scripts into a reusable, discoverable service layer.
• **Self‑learning:** The built‑in teaching protocol records successful tool‑calls and feeds them back into the model context, improving agent performance over time without manual prompt engineering.

**Practical Adoption Path**

1. **Prototype:** Drop the squeez binary (or add the Rust crate) into an existing CI pipeline, configure the MCP endpoint for the target AI CLI, and enable the --compress flag.
2. **Tool‑binding:** Register your internal tools (e.g., deployment scripts, data‑fetchers) via the provided hook manifest; the system will automatically deduplicate repeated calls.
3. **Production rollout:** Deploy the MCP server behind a reverse proxy, point all AI‑assistant instances to it, and enable the signature‑mode for read‑only code inspections.
4. **Iterate:** Use the self‑teaching logs to refine hook definitions and optionally feed the learned context back into custom model fine‑tuning.

**Production Readiness**

• **Activity & adoption:** 160 ★, 15 forks, recent commits (as of 2026‑06‑30) and usage across multiple AI CLI projects indicate an active community.
• **Technical maturity:** Zero runtime dependencies, a single compiled binary, and a well‑documented Rust API reduce operational overhead.
• **Risk considerations:** License and security posture still need a final audit, and long‑term maintainership should be confirmed, but the current signals are strong enough for a serious pilot in a controlled environment.

Overall, squeez offers a low‑friction, high‑impact way to make AI agents more cost‑effective and tool‑aware, making it a viable candidate for production‑grade deployments after the standard compliance checks.

Problem

claudioemmanuel/squeez helps connect AI assistants to real tools and data through a standard protocol.

Outcome

High for an OSS candidate: recent activity, adoption, and ecosystem signals are strong enough for a serious pilot.

Workflow

• connect AI agents to tools
• ship Model Context Protocol servers
• standardize integrations

Evidence

Hook-based token compressor for 5 AI CLI hosts (Claude Code, Copilot CLI, OpenCode, Gemini CLI, Codex CLI). Up to 95% bash compression, signature-mode for code reads, cross-call dedup, MCP server, self-teaching protocol. Zero runtime deps.

Source