RepoPicks Radar

Updated every day around 02:00 UTC

Guides

Radar home

Caveman

JuliusBrussee/cavemanReviewed by AI

What it does

A set of add-ons for AI coding assistants such as Claude Code, Codex, Cursor and Gemini CLI. It installs instructions that make the assistant answer in a clipped style, plus a local proxy that compresses logs, JSON, CSV and web pages before the model reads them, and an optional router that picks a model per question. You install it in a terminal with one npx command and Node 22.13 or newer; it then edits the settings of each assistant it finds on your machine.

How easy it is to install

C · install 1 or 2 tools · Windows

  • Windows C
  • macOS C
  • Linux C

C on all three: you need Node.js 22.13 or newer and a terminal, then `npx @caveman-ai/cli@latest`, which detects your assistants and changes their configuration. Windows needs PowerShell 5.1 or newer for the one-line form; the released files are command-line binaries, not an app you open.

Numbers on GitHub

Stars
116,527on Oct 31, 2026
Downloads
722,81884,456 this week
Growth
+1,216stars in 5 days

Stars over time

115,000115,500116,000116,500117,000 Oct 26Oct 27Oct 28Oct 29Oct 30Oct 31 115,000115,500116,000116,500117,000 Oct 26Oct 29

Facts

Runs on
Windows, macOS, Linux
You need to install
Node.js
To build from source
Node.js, Docker, Go
Disk space
46 MB–137 MB · from the installer
Accounts or keys
Caveman Cloud account (for the Auto router), Your own model provider key (Anthropic, OpenAI and others)
Cost
Freemium · Apache-2.0 code, free to run. The "Auto" router needs a free Caveman account, and Caveman Cloud is a separate paid service (app.caveman.so).
License
Apache-2.0 (Permissive)
Can replace
OpenRouter, Portkey
Last update
Oct 25, 2026

Risk signals (7)

  • Attentionunusual star rate

    116532 stars in 210 days (~555 per day)

    automatic check

  • Attentiontelemetry on by default

    README Privacy section: "The CLI sends usage stats by default: commands run, token counts, caveman-blocks commands and usage counts, a random install ID, OS, and IP", and says it is turned off with `caveman telemetry off` or `DO_NOT_TRACK=1`.

    AI review

  • Attentionprompts sent to vendor cloud when routing on

    README Privacy section on Auto: "each ask sends Caveman Cloud your latest ask (with whatever your agent attaches to it, such as CLAUDE.md, memory or @-mentioned files) ... On the Free plan Caveman may keep it to improve routing", and a request routed to a Caveman Cloud model "passes through Caveman Cloud, whole (system prompt, tool results, code)".

    AI review

  • Attentioninstaller rewrites other tools settings

    INSTALL.md: the installer "Auto-detects every supported agent installed on your machine", "runs that agent's native install path (plugin / extension / rule file / `npx skills add`)" and "Wires Claude Code hooks and statusline badge on top, plus a Codex start-of-session hook" — it changes files and hooks belonging to around 30 other programs.

    AI review

  • Infoinstall script piped from web

    INSTALL.md offers `curl -fsSL https://raw.githubusercontent.com/JuliusBrussee/caveman/v4.0.1/install.sh | bash` and `irm ...install.ps1 | iex`, and warns in its own words: "Piping a script straight into a shell runs it sight-unseen." The scripts then run `npx -y github:JuliusBrussee/caveman#v4.0.1`, pinned to a tag the owner can move.

    AI review

  • Infoself reported and partial benchmarks

    The README's own caveats: the 33.2% session figure "is from August 2026 on an older engine", HTML came out "9.9% worse", the Auto comparison says "the run is half done" and the score gap "is too small to tell apart at this size", and the Blocks part is "Pre-release (v0.1.0-rc.5). Python only, no sandbox yet."

    AI review

  • Inforelease binaries unsigned by github

    Only the bin-v4.0.2 release carries binaries (caveman-proxy, caveman-engine, caveman-mcp, cavemem and others for all three systems); INSTALL.md says runtime binaries are "checked against a checksum list signed with a key built into the CLI", which cannot be verified from outside.

    AI review

Evidence is shown in its original language.

What Radar could not check

  • Whether the token savings hold on your own projects and models
  • Whether the outside citations and blog results quoted in the README say what the README says they say
  • What exactly the installer changed in each assistant's configuration, and whether uninstalling restores it
  • What the telemetry and the Auto router send once running, since that needs watching the traffic

Similar projects

The RepoPicks app

Coming soon: install in 1 click with the RepoPicks app.