Computer-Use Cache CLI
A local OpenAI-compatible proxy that caches repeated agent requests and replays stored answers without calling the model.
- 52 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- The proxy sees the upstream API key and all prompt and response traffic.
- The cache stores request and response content on disk.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add cli/computer-use-cacheDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The proxy sees the upstream API key and all prompt and response traffic.
- The cache stores request and response content on disk.
This is third-party code. Review the repository files before installing.
What it does
Computer-Use Cache sits between your app and any OpenAI-compatible provider. Misses go upstream, successful JSON responses are stored and exactly repeated requests are served from cache. The idea is that computer-use agents keep repeating the same planning, browser and coding steps and paying twice for them is wasteful. It installs as a zero-dependency npm package or as a Python server with SQLite. The `computer-use-cache` command supports `start`, `install`, `init`, `stats`, `clear` and `env`. Install configures Codex, Claude Code, Cursor, OpenClaw and Hermes to call the local address. The cache lives in a file or SQLite, lifetime and model allow and deny lists are configurable, and responses carry hit and miss headers.
Who it is for. Agent developers and teams whose workflows repeat and who need a transparent cache in front of OpenRouter or OpenAI.
Good fit when
- The same agent scenario runs many times during debugging.
- You need a local cache in front of OpenRouter with no client code change.
- Claude Code, Codex and Cursor must be pointed at one local address.
- You need to watch the cache hit share through a statistics command.
Not a fit when
- Requests differ every time and never repeat.
- A fresh model answer is required for every call.
- Prompts and responses cannot be stored on disk.
Example request
Start computer-use-cache in front of OpenRouter, point Claude Code at the local address and then show the hit statistics.Limitations
The cache stores prompts and responses on disk and the proxy sees your upstream API key and all traffic, so run it on a trusted machine and restrict access to the cache directory. Only exactly matching requests are served, and similar requests trigger a new call. A stale answer may return until the lifetime expires, so shorten it for tasks with changing data. The README states no app authentication, billing or product state, as it is a deliberately small proxy. The benchmark results in the README were produced by the author.
How to disable. Stop the proxy, remove the settings added by `install` and restore the previous `OPENAI_BASE_URL` in the clients. Clear the cache directory with `clear` or by hand.
FAQ
How do I run the proxy?
Set `UPSTREAM_BASE_URL` and `UPSTREAM_API_KEY`, run `npx -y github:rohanarun/computer-use-cache install all`, then `npx -y github:rohanarun/computer-use-cache start`.
How do I point a client at the cache?
Set `OPENAI_BASE_URL=http://127.0.0.1:8000/v1` and use your upstream key as `OPENAI_API_KEY`.
Which providers work?
OpenRouter by default and OpenAI directly through `UPSTREAM_BASE_URL`. Any provider with a compatible API works.
Which agents does install configure?
Codex, Claude Code, Cursor, OpenClaw and Hermes. They can be installed one at a time: `install codex`, `install cursor` and so on.
How do the npm and Python versions differ?
The npm version has zero dependencies and a file-backed cache, while the Python version uses SQLite with a TTL and model lists.
Discussion
No questions or reviews yet.
Related
CLI toolsParitok is a proxy between a coding agent and the model API that compresses requests and cuts the input-token bill.
Local proxy between agent and model API: compresses diffs and logs, keeps the request prefix stable for prompt caching.
A long-term memory MCP server for agents: one Rust binary that runs offline and never calls a language model
A local Go agent that records runs, verifies them offline by hash, forks with another model and controls the browser and macOS
Local memory for agents in a single Rust binary: stores, searches by meaning and keyword, works with no network and no keys
A proxy between an agent and a local model: old tool outputs go to a disk archive the model reads back with a recall tool
README in short
The proxy replays requests from cache and forwards misses upstream. The README covers setup for five agents and commands for stats, clearing and printing environment variables.