MrGray17 / opentoken

What is opentoken?

OpenToken is a plugin for OpenCode that slashes the tokens you burn through during coding sessions. It intercepts tool output before it reaches the LLM and runs it through 35 layers of compression -- everything from stripping ANSI escape codes to replacing entire repeated code blocks with tiny markers. The model still sees the same semantic meaning but talks completely normally. It also compresses model responses on the way out. The plugin claims 74% overall compression in production, with 93% median on compressible calls.

Why is it gaining traction?

The hook is simple: AI coding costs money, and this openly saves tokens without making the model sound like a caveman. Most compression tools degrade output quality -- OpenToken's "zero-risk principle" means every stage only applies if it actually shrinks things. It handles seven tool families specifically (git, npm, cargo, docker, pip, make, and filesystem), plus generic text and JSON. The LSP-first enforcement is interesting: it blocks symbol searches through grep and suggests real LSP tools instead, cutting navigation costs by 80%. It runs on Bun with zero config upfront.

Who should use this?

OpenCode power users burning through API quotas who want to extend their budget. Teams working on large codebases where every grep or read call generates massive output. Developers in token-heavy languages (verbose TypeScript configs, sprawling JSON) who notice their context windows filling up fast. If you're just dabbling in OpenCode, the overhead of yet another plugin probably isn't worth it -- but if you're running it daily on production projects, the savings compound quickly.

Verdict

This is a clever, well-architected tool with proven real-world numbers behind it. At 34 stars with a 1.0% credibility score, it's early and unproven at scale -- the README metrics look real but aren't independently audited. The TypeScript codebase is substantial and the test coverage exists, but you're adopting a hobby project for something mission-adjacent to your workflow. Use it if you're already deep on OpenCode and want to experiment with token costs, but don't bet critical work on it staying stable as the author iterates.