Status: [LIVE] Endpoints found: https://pypi.org/project/aip-proxy Source: https://github.com/futesat/tokenlean # tokenlean [![Docker Tests](https://github.com/futesat/tokenlean/actions/workflows/docker-tests.yml/badge.svg)](https://github.com/futesat/tokenlean/actions/workflows/docker-tests.yml) **tokenlean** chains [aip-proxy](https://pypi.org/project/aip-proxy/) and [LiteLLM](https://github.com/BerriAI/litellm) in front of the GitHub Copilot API so that Claude Code (or any OpenAI-compatible tool) can use your Copilot subscription as a backend. It also integrates [rtk](https://github.com/rtk-ai/rtk) as a Claude Code hook to compress command outputs, delivering two independent layers of token reduction. > **Double savings** > | Layer | What it compresses | Reduction | > |---|---|---| > | **[rtk](https://github.com/rtk-ai/rtk)** | Shell command *outputs* (git, ls, grep…) before they reach the model context | 60–90% | > | **[aip-proxy](https://pypi.org/project/aip-proxy/)** | Input *prompts* (whitespace, comments, deduplication) | 15–40% | ## How it works ``` Your app / tool │ ▼ HTTP :4444 aip-proxy ← compresses prompts 15-40% (whitespace, comments, deduplication) │ ▼ HTTP :4445 LiteLLM ← translates OpenAI API calls to GitHub Copilot API │ ▼ HTTPS GitHub Copilot API ``` ## Available models All models configured in `copilot-config.yaml`. Use the `model_name` value as the `model` field in your API calls. ### OpenAI | Model name | Underlying model | Reasoning effort | | --------------------- | ------------------ | ---------------- | | `gpt-4-1` | GPT-4.1 | — | | `gpt-4o` | GPT-4o | — | | `gpt-5-mini` | GPT-5 mini | high | | `gpt-5-1` | GPT-5.1 | high | | `gpt-5-1-codex` | GPT-5.1 Codex | high | | `gpt-5-1-codex-max` | GPT-5.1 Codex Max | xhigh | | `gpt-5-1-codex-mini` | GPT-5.1 Codex mini | high