llms.txt · git:20260518.53d8fd8 · 2026-05-18 · sha256 6ca285531a7185a9
llms.txt git:20260518.53d8fd8A
Immutable. This exact content never changes and is served at /api/v1/blob/6ca285531a7185a9.
# Caveman Code — terminal coding agent > Caveman Code is a terminal coding agent that uses 2× fewer tokens than > Codex CLI on identical tasks, supports 20+ providers with 6 OAuth flows, > and ships under the MIT license. Four-layer Caveman Mode compression > gives ~85% reduction on tool output. Plan mode, autonomous goal loop, > worktree-isolated subagents, native MCP, cavemem-backed memory. This file is the LLM-friendly entry point to Caveman Code's documentation. Each linked page is plain markdown. ## Install ``` npm install -g @juliusbrussee/caveman-code ``` Requires Node.js 20+. Binaries `caveman` (primary) and `caveman-code` (alias) are both registered on PATH. ## npm package - Main CLI: `@juliusbrussee/caveman-code` - Sub-packages (auto-installed): - `@juliusbrussee/caveman-agent` - `@juliusbrussee/caveman-ai` - `@juliusbrussee/caveman-tui` - `@juliusbrussee/caveman-sdk` - `@juliusbrussee/caveman-markdown-preview` - Optional surfaces: `@juliusbrussee/caveman-mom`, `@juliusbrussee/caveman-pods`, `@juliusbrussee/caveman-web-ui` ## Top-level docs - [Quickstart](docs/getting-started/quickstart.md): zero-to-first-prompt in 30 seconds - [Install](docs/getting-started/installation.md): macOS, Linux, Windows, Docker, Homebrew, npm - [Auth & Providers](docs/getting-started/auth.md): 20+ providers, 6 OAuth flows - [Models](docs/getting-started/models.md): default models, thinking levels, architect/editor split ## Reference - [Slash Commands](docs/reference/slash-commands.md): built-ins + authoring markdown commands - [Recipes](docs/reference/recipes.md): YAML pipelines, Goose-style schema - [MCP](docs/reference/mcp.md): client + cave-as-server, stdio/HTTP/inproc transports - [Daemon](docs/reference/daemon.md): caveman serve, multi-client attach, worker mode ## Migration - [From Claude Code](docs/migration/from-claude-code.md): paste your config, it works - [From Codex](docs/migration/from-codex.md): AGENTS.md, .mcp.json, sandbox parity - [From Aider](docs/migration/from-aider.md): repo map and edit-format-per-model ## Cookbook & API - [Cookbook](docs/cookbook.md): GitHub Actions, multi-agent review, daemon pair programming - [Comparison](docs/comparison.md): Caveman Code vs Claude Code / Codex / Aider / Crush / opencode - [Troubleshooting](docs/troubleshooting.md): install, auth, sessions, tools, hooks, MCP - [API](docs/api.md): SDK, JSON-RPC, OpenAPI, extension API ## Differentiators (one-line each) 1. Caveman Mode compression: 4 layers, ~85% reduction on tool output 2. 20+ providers + 6 OAuth flows (Claude Pro, ChatGPT Plus, Copilot, Gemini, Antigravity) 3. Session branching (`/tree`, `/fork`) 4. Plan mode + autonomous goal loop 5. Worktree-isolated subagents via Task tool 6. Native MCP client; caveman-code can serve as MCP server 7. Aider-grade repo map (PageRank) + edit-format-per-model 8. Daemon (`caveman serve`) survives SSH drops; multi-client attach 9. Shadow-git checkpoints + `/rollback N` 10. cavemem-backed memory with episodic-to-semantic consolidation 11. Claude Code authoring formats (settings.json, commands, skills, agents) — paste-and-go compat 12. MIT licensed; no telemetry by default ## Repo - Source: https://github.com/JuliusBrussee/caveman-cli - Issues: https://github.com/JuliusBrussee/caveman-cli/issues - Releases: https://github.com/JuliusBrussee/caveman-cli/releases - npm: https://www.npmjs.com/package/@juliusbrussee/caveman-code