GraphCanon
Home/Developer Tools/caveman
caveman logo

caveman

JuliusBrussee/caveman

Reduce token usage with concise 'caveman'-style prompts.

GitHub Website

GraphCanon updated 5d · GitHub synced 5d · 37 views this month

98k stars5.7k forksLast push 5d Go MIT

Decision brief

The **caveman** tool is designed for developers and AI users who aim to optimize their token usage through the generation of more concise prompts, thereby potentially reducing costs and improving efficiency. However, it犺

Good fit when

  • When you need to significantly cut down on token usage in AI interactions, up to 65%, without losing essential information content.
  • If your project can benefit from a simplified or direct language style that still conveys necessary instructions effectively.

Avoid when

  • When requiring complex and detailed prompts that necessitate more nuanced expression beyond simple, 'caveman'-style sentences.
  • For situations where adherence to formal or specific linguistic structures is mandatory for the task's success.

Observed Jul 11, 2026 · Source: enrich:decision_facts

caveman alternativesTrust report
caveman vs ponytailcaveman vs tooncaveman vs ECC

Verify the decision

Maintenance and security

Full trust report
Maintenance
Very active (0d since push)
As of 5d
Provenance
Not a fork · Personal account
As of 5d
Security (OSV)
No lockfile
As of 1mo

Public GitHub metadata and optional OSV scans. Signals, not a guarantee. Trust methodology.

Install

go get github.com/JuliusBrussee/caveman
pkg.go.dev

How it fits your stack(15)

Typed graph edges - alternatives, integrations, successors, and dependencies. Ranked by relationship type, not raw GitHub stars.

Alternative

Relationship graph

Optional deeper exploration of typed edges and category neighbours.

Similar tools

Same-category neighbours not already linked as typed edges.

Evidence and technical details

Sourced facts, taxonomy, compatibility claims, README excerpt, and machine-readable endpoints.

Overview

A JavaScript-based repository offering a Claude Code skill that minimizes the number of tokens used in AI interactions by employing simplified language structures, reducing token count by up to 65%.

Capability facts

CLI
CLI entrypoint

Source: package.json:bin|scripts · Aug 16, 2026

MCP server
No MCP server detected

Source: repo_scan · Aug 16, 2026

Languages
go, javascript

Source: github.language+package.json · Aug 16, 2026

Categories

Developer ToolsLLM Frameworks

Compatibility

Sourced claims from the README excerpt - not unsourced marketing copy.

Node.js runtimeNode.js

Source: README excerpt (regex_v1, Aug 16, 2026)

The skill installer needs Node.js 18+, finds supported agents already on your machine, skips the rest, and is saf
Source link
Python runtimePython

Source: README excerpt (regex_v1, Aug 16, 2026)

**MIT** — the skill, Agent SDK and initializer, the CLI, both client SDKs (TS + Python), kit, evals/graders, contracts, provider catalog, the extension shell, and the
Source link

Tags

aianthropiccavemanclaude-codeprompt-engineeringtokens

README

Install

Caveman is three separate installs. Each works alone. Not sure? Start with the skill — it needs no account, proxy, Go toolchain, or code changes.

Shorter answers — the skill (MIT). The caveman skill + slash commands in every supported agent found on your machine. Nothing else.

npx skills add JuliusBrussee/caveman

Smaller inputs — Caveman Proxy (BSL-1.1 runtime; MIT CLI). The caveman CLI plus signed local binaries: proxy, engine, MCP recovery, memory — and the optional browse + shrink tools. Then caveman claude wraps your agent.

npm install -g @caveman-ai/cli && caveman setup --install

Compressed browsing — Caveman Browse (BSL-1.1). Included in caveman setup --install (needs Chrome) — a local Chrome driver your agent reaches as MCP tools.

caveman browse <url>

A new agent — Agent SDK. A TypeScript agent project on the native Caveman runtime. Client SDKs: npm i @caveman-ai/sdk · pip install caveman-sdk.

npm create @caveman-ai/agent@latest my-agent

On Windows, the skill installs with irm https://raw.githubusercontent.com/JuliusBrussee/caveman/v1.10.0/install.ps1 | iex (PowerShell 5.1+).

The skill installer needs Node.js 18+, finds supported agents already on your machine, skips the rest, and is safe to rerun. Prefer one agent only?


---

### Install the CLI

```bash
npm install -g @caveman-ai/cli
caveman setup --install   # downloads the signed runtime binaries

setup --install verifies the signed checksum manifest and the SHA-256 of every binary before an atomic install. Prefer building from source? A clone plus scripts/install-local-cli.sh (macOS/Linux) or pwsh -File scripts/install-local-cli.ps1 (Windows) still works — that path needs Go and pnpm. SDK users can point provider base URLs at the local Proxy directly (ANTHROPIC_BASE_URL=http://127.0.0.1:8787/anthropic).

caveman claude                  # full stack (default): S4 compress + TOON best-of + caveman & browse MCP tools + output shrink
caveman wrap --off codex        # byte-safe pass-through metering only
caveman wrap --pixel claude     # lossy text → PNG pixel mode (model-gated)

Subscription logins work — see the note below.

ModeWhat it doesBytes the model sees
default stack (caveman claude)Structural compression routed per content type (table below), plus uniform JSON tool results re-encoded as TOON only when measured smaller; config toon: false turns it off.Changed, recoverable
--offCounts tokens and cost. Changes nothing.Byte-identical
--pixelDense text slabs rendered to PNG pages for vision models.Changed, recoverable

Safety gates stay explicit:

  • CCR first. Before a lossy transform goes upstream, original bytes land in CCR, a content-addressed store on your disk. The agent retrieves them through caveman_retrieve or caveman retrieve <handle>. Parse problem, store failure, or larger result sends original bytes unchanged.
  • Visible declines. Pixel refuses sparse code. Convert refuses skills when PNG pages do not beat text. TOON runs only when its output measures smaller. Each decline includes its reason.
  • Labeled evidence. Local results report inferred: estimates for choosing what to try. verified requires real traffic and eval gates. Offline caveman never reports it.

License

Split license. Skill and adoption surfaces are MIT. Engine-linked runtime is BSL-1.1 source-available, not OSI Open Source before Change Date.

MIT — the skill, Agent SDK and initializer, the CLI, both client SDKs (TS + Python), kit, evals/graders, contracts, provider catalog, the extension shell, and the thin cavemem clients.

BSL-1.1 — Engine, Proxy, Cache Engine, rewriter, Browse, MCP server, shrink, cavemem Go core, and shared Go platform. New Engine-linked runtime modules default to BSL-1.1. Source-available: read

For agents

This page has a .md twin and JSON over the API.

/tools/juliusbrussee-caveman.mdGET /api/graphcanon/tools/juliusbrussee-cavemanGET /api/graphcanon/graph?tool=juliusbrussee-caveman

Was this helpful?

Anonymous feedback helps us improve pages and translations.