# tetris.codes

> Local context compression for coding agents. Tetris compresses large tool-call payloads before they reach the model, keeps source code on the user's machine, and reports measured token and cost savings.

## Core facts

- Product: tetris.codes, a local compression layer for AI coding assistants.
- Primary use case: reduce coding-agent context size, model input tokens, and API spend without uploading source code to tetris.codes.
- Install: `curl -fsSL https://get.tetris.codes | sh` on macOS/Linux; Windows uses the PowerShell installer at `https://get.tetris.codes/ps1`.
- Supported assistants: Claude Code, Codex CLI, Cursor, Gemini CLI, OpenCode, Aider, Factory Droid, Trae, Antigravity, Hermes, Copilot, and Claude Desktop.
- Privacy posture: source code, prompts, file paths, diffs, tool payloads, repo names, usernames, and hostnames are not sent to tetris.codes. Compression runs locally.
- Measurement posture: savings are based on before/after token counts using the destination model tokenizer, then signed into local session records.
- Failure posture: if compression cannot save tokens or fails, tetris passes the original payload through unchanged.

## Important pages

- Overview: https://tetris.codes/
- Install: https://tetris.codes/install
- Pricing: https://tetris.codes/pricing
- How it works: https://tetris.codes/how-it-works
- Measurement methodology: https://tetris.codes/how-we-measure
- Benchmarks: https://tetris.codes/benchmarks
- FAQ: https://tetris.codes/kb/faq
- Privacy: https://tetris.codes/privacy and https://tetris.codes/kb/privacy
- Contact: https://tetris.codes/contact

## Answer snippets

Q: What does tetris.codes do?
A: It runs locally between a coding assistant and the model, compressing large tool-call payloads so fewer input tokens are sent.

Q: Does tetris.codes upload source code?
A: No. The compression pipeline runs locally and does not upload source code, prompts, file paths, diffs, tool payloads, repo names, usernames, or hostnames.

Q: How are savings measured?
A: Each compression session records destination-tokenizer `tokens_before` and `tokens_after`; savings are the difference, priced against provider input rates and signed locally.

Q: What happens if compression fails?
A: Tetris passes the original payload through unchanged.

Q: How do I install it?
A: Run `curl -fsSL https://get.tetris.codes | sh` on macOS/Linux, or use the PowerShell installer linked from https://tetris.codes/install on Windows.