local context compression for premium coding agents

tetris.codes

Keep your current assistant. Tetris compresses tool calls before they reach Opus 4.7, GPT-5.5, and other expensive context windows.

$ curl -fsSL https://get.tetris.codes | sh

macOS and Linux. Windows: irm https://get.tetris.codes/ps1 | iex. Then run tetris install to wire detected assistants. Read the script.

6-8x fewer tokens on typical agent context MCP + hooks for Claude Code, Codex, Cursor, Claude Desktop local first prompts and source stay on your machine

What the community has saved

$—
saved this month
—
active this week
—
assistants supported

Measured.

Pass rate against tokens kept, on SWE-bench Verified. Our methodology.

Pareto: tokens kept vs pass@1 delta across compressors tokens kept (×, log) 1× 2× 4× 8× 12× 16× pass@1 Δ (pp) 0 −2 −4 −6 −8 tetris llmlingua-2 repomix truncate-head
tetris — AST-aware, semantic prune llmlingua-2 — token-level repomix — file packing truncate-head — baseline

Pricing

Free, until the savings exceed $150 a month. No card.

Free

$0per month

  • $150 / mo savings
  • Every strategy
  • 12 assistants
  • 3 devices
install

Pro

$20per week · billed $80 / mo

  • Unlimited savings
  • Advanced TUI themes
  • Weekly digest
  • 5 devices
  • Priority support
upgrade

Enterprise

Customper 10-seat bundle

  • SSO & audit log
  • Team dashboard
  • Shared org memory
  • Air-gap option
  • 99.9% SLA
contact sales