Paritok

Paritok is a compression gateway that sits between AI coding agents and their LLM backend. It intercepts requests and non-destructively compresses tool schemas, file reads, and conversation history before they reach the model. The result is smaller token bills and longer sessions without hitting context limits. You can self-host it for free or use the cloud version with an API key.

How it works: you set one environment variable (ANTHROPIC_BASE_URL) to point to Paritok, and it starts rewriting requests. It uses a 4B-parameter compression model trained on 45,000 real agent trajectories to decide what to keep and what to summarize. Tool schemas are stubbed to only include relevant ones, file reads are compressed while preserving identifiers and error messages, and stale history is summarized once a budget is reached. Nothing is lost — the agent can request original bytes locally without burning a turn. The page claims up to 85% token reduction in context-saturated sessions.

Who benefits: developers and teams using AI coding agents like Claude Code, Cursor, Codex, or OpenHands. If you're paying for LLM API calls and running long coding sessions with many tool calls, Paritok can cut your input bill significantly. The self-hosted option is free and open-source under Apache-2.0, making it accessible for any team.