🪨 Caveman Code

The terminal coding agent that talks like a caveman — and burns half the tokens doing it.

Same model. Same task. ~2× fewer tokens than Codex. 20+ providers · plan mode · autopilot loop · MIT.

Install · The Trick · How It Saves Tokens · Why Caveman · Features · SDK

---

🔥 The trick

Big agent waffle. Waffle cost token. Caveman no waffle.

Asked ▸ why does this component re-render on every keystroke?

Ordinary agent · ~290 tokens	🪨 Caveman Code · 31 tokens
Great question! A React component can re-render on every keystroke for several reasons. The most common cause is passing a fresh object or function reference as a prop on each render, which defeats React's referential-equality bail-out and forces the child to reconcile again … (three more paragraphs)	New object ref each render. Inline prop = new ref = re-render. Wrap in useMemo.

Same answer. Same model. Caveman version costs ~9× less to read back — and the agent reads its own context back on every single turn. The saving compounds across the whole session.

That is the entire product. Everything below is the coding agent it ships inside.

---

The proof

25-task MicroBench · gpt-5.5 · xhigh reasoning · 2026-05-18

Tool	Fresh tokens	Pass rate
🪨 caveman	████████████▌ 524k	14 / 25
codex	████████████████████████ 1,010k	15 / 25

1.93× fewer tokens than Codex CLI on identical tasks. Same gpt-5.5 model. Same xhigh reasoning. Pass rate within one task.

No marketing-deck baselines. Each tool spawned as a real child process. Each task verified by a task-specific verify.sh. Raw CSV + per-task logs published.

npx tsx research/evals/run-honest-bench.ts --tools caveman,codex   # reproduce in one command

Raw CSV · Aggregate JSON · Methodology · 25 task prompts

---

Install

npm install -g @juliusbrussee/caveman-code

Installs two binaries — caveman (primary) and caveman-code (alias). Same command, pick either.

export ANTHROPIC_API_KEY=sk-ant-...     # or any supported provider's key
caveman                                 # launch the TUI
caveman "explain this codebase"          # one-shot
caveman --goal "ship feature X"          # autonomous loop

Other install paths — pnpm · yarn · bun · Docker · OAuth login

pnpm add -g @juliusbrussee/caveman-code
yarn global add @juliusbrussee/caveman-code
bun  add -g @juliusbrussee/caveman-code

# Docker
docker run --rm -it -v "$PWD:/work" ghcr.io/juliusbrussee/caveman-code:latest

# No API key? Use a subscription you already pay for:
caveman && /login   # Claude Pro · ChatGPT Plus · Copilot · Gemini · Antigravity

CI / headless install: docs/getting-started/installation.md.

---

Quick Start

caveman                            # interactive TUI
caveman "fix the failing tests"     # start with a prompt
caveman -p "summarize this file"    # non-interactive: print and exit
cat err.log | caveman -p "debug"    # pipe stdin
caveman -c                          # continue last session
caveman -r                          # browse + resume sessions
caveman --plan "ship payments v2"   # plan mode — read-only
caveman --goal "ship payments v2"   # autonomous Ralph loop

Type / inside the TUI for every slash command. Reference: docs/reference/slash-commands.md.

---

How It Saves Tokens

Four compression layers, always on — and they hit two separate token sinks: what the model says and what the shell returns.

Token sink	Layer	What happens	Cut
Model reply	Caveman Mode	Terse technical fragments — no filler, no hedging. Levels lite · full · ultra.	prompt + reply
Tool output	Tool Budgets	Per-tool line caps (bash 80 · read 300 · grep 120), ANSI strip, blank-line collapse, semantic JSON/XML extraction.	−67% to −94%
	Read Dedup	Files fingerprinted per session — re-reads return a stub, not the bytes.	−99% on repeats
	RTK	Optional external Rust binary ("Rust Token Killer") — pipes bash output through rtk before it enters context.	−60% to −90% (RTK's own bench)

Pays for itself after one tool call.

Benchmark — 10 real tool-output fixtures · −86% aggregate

  git diff (901 lines)   ██████████████████████████████████████████████████  -94%
  npm ls (701 lines)     ████████████████████████████████████████████████    -92%
  ls recursive (601 ln)  ███████████████████████████████████████████████     -90%
  grep results (801 ln)  █████████████████████████████████████████████       -89%
  test output (501 ln)   ████████████████████████████████████████████        -88%
  XML/pom.xml (382 ln)   ████████████████████████████████████████            -79%
  docker inspect (258)   ██████████████████████████████████                  -68%
  ANSI colored (97 ln)   █████████████████████████████                       -50%
  read file (429 lines)  ████████████████                                    -32%
  build output (19 ln)   █████████                                           -18%
                         ────────────────────────────────────────────────────
  AGGREGATE              ███████████████████████████████████████████████     -86%

Metric	Value
Tokens saved (10 fixtures)	~72,400 of 337K chars
System-prompt overhead	120–195 tokens (lite–ultra)
Net savings — 15-turn session	+567K tokens (~$1.70, Sonnet)
Net savings — 30-turn session	+1.13M tokens (~$6.92, Sonnet)

npm run bench:offline   # compression analysis — free, <1s
npm run bench:replay    # analyze your real sessions — free
npm run bench:live      # A/B with live LLM calls — needs API key

caveman --caveman-mode ultra   # most aggressive
caveman --caveman-mode lite    # system-prompt compression only
caveman --no-caveman-mode      # off

---

Why Caveman Code

Capability	Caveman	Claude Code	Codex	Aider	opencode
4-layer token compression	✅	❌	❌	repo map only	❌
20+ provider OAuth	✅	Anthropic	ChatGPT	API keys	✅
Autonomous goal loop	✅	❌	❌	❌	❌
Autopilot — no permission prompts	✅	❌	❌	✅	❌
Repo map (PageRank, Aider-style)	✅	❌	❌	✅	❌
Architect / editor model split	✅	❌	❌	✅	❌
Session branching + shadow-git checkpoints	✅	❌	fork only	git only	❌
Persistent semantic memory (cavemem)	✅	MEMORY.md	❌	❌	❌
MIT open source	✅	closed	Apache-2.0	Apache-2.0	✅

Full table including Crush: docs/comparison.md.

---

Features

	Feature	Trigger
🤖	Autonomous goal loop — Ralph-style autopilot. Rolling state, per-iteration $/token ledger, shadow-git checkpoints, ranked termination (sentinel · iteration cap · $-cap · no-progress · SIGINT). Resume any time.	caveman --goal
🧠	Plan mode — read-only chat. Model sees only read/grep/find/ls, produces a written plan, never edits. Subagents inherit the gate. /act to execute.	--plan · /plan
👥	Subagents — up to 7 parallel, worktree-isolated. Frontmatter agents at .cave/agents/*.md (Claude Code superset). Five ship by default.	Task tool
🪞	Architect / editor split — slow model plans, fast model executes. ~3–5× cheaper than a single-model run.	--architect · --editor

Latest release: plan mode · goal loop · native memory tools · subagent registry. Full history → CHANGELOG.md.

More — sessions · providers · MCP · memory · recipes · daemon · CLI flags

🌳 Sessions, branching, replay

JSONL sessions in ~/.cave/agent/sessions/, organized by working directory. Branching never overwrites history.

caveman -c                    # continue most recent
caveman -r                    # browse and select
caveman --fork <path|id>      # fork into a new file

/tree navigate + branch in-place (search · fold · page · filter) · /compact manual compaction · /checkpoint + /rollback N rewind code and conversation together.

🌐 20+ providers, 6 OAuth flows

OAuth — Claude Pro/Max · ChatGPT Plus/Pro · GitHub Copilot · Google Gemini · Antigravity · Vertex API keys — Anthropic · OpenAI · Azure · Vertex · Bedrock · Mistral · Groq · Cerebras · xAI · OpenRouter · Vercel AI Gateway · Hugging Face · Kimi · MiniMax · Z.AI · DeepSeek Custom — any OpenAI-/Anthropic-/Google-compatible endpoint via ~/.cave/agent/models.json.

🔌 MCP, hooks, skills, commands — Claude Code-compatible

Authoring formats are a superset of Claude Code's — paste your existing config, it works.

Claude Code	Caveman	Notes
~/.claude/settings.json	~/.cave/settings.json	Hooks identical (run as observers, never block)
~/.claude/commands/*.md	~/.cave/commands/*.md	Frontmatter superset
~/.claude/skills/<name>/SKILL.md	~/.cave/skills/<name>/SKILL.md	Identical
~/.claude/agents/<name>.md	~/.cave/agents/<name>.md	Frontmatter superset
.mcp.json	.mcp.json	Same path, no change

MCP transports: stdio · Streamable HTTP · in-process. OAuth 2.1 + PKCE; tokens in OS keychain.

caveman mcp add <name>      # add a server
caveman mcp doctor          # health-check + tool listing
caveman mcp-server          # run caveman itself as an MCP server (Codex-compatible)

🧠 Memory via cavemem

Persistent memory delegated to cavemem (MIT, hybrid BM25 + local vectors). Agent has two native tools — memory_search and memory_save; relevant recall is auto-injected each turn.

/memory search "auth migration"
/memory consolidate            # cluster recent observations into semantic facts
/memory sync --from claude     # import Claude Code's MEMORY.md

🛠️ Recipes

Declarative multi-step YAML workflows at ~/.cave/recipes/*.yaml. Ten built in: accessibility-audit · add-feature-flag · add-tests · bump-deps · extract-component · migrate-deps · migrate-to-biome · port-to-typescript · release · seo-audit.

/recipe run add-tests src/auth.ts

🖥️ Daemon

caveman serve --port 39245             # start the daemon
caveman attach --host localhost:39245  # attach a TUI

Sessions live in SQLite and survive SSH drops. Prepend & to any prompt to dispatch to a remote caveman worker.

⚙️ CLI flags

Flag	Description
-c / -r	Continue / browse-resume session
-p, --print	Non-interactive: print and exit
--mode json|rpc	Structured output modes
--plan / --goal <prompt>	Plan mode / autonomous loop
--provider / --model	Provider name / model ID (:<thinking> suffix ok)
--thinking <level>	off·minimal·low·medium·high·xhigh
--architect / --editor <model>	Architect/editor split
--caveman-mode <level> / --no-caveman-mode	lite·full·ultra / off

Env: ANTHROPIC_API_KEY · OPENAI_API_KEY · CAVE_CODING_AGENT_DIR (config dir) · CAVE_CACHE_RETENTION=long (extended prompt cache).

---

SDK

import { AuthStorage, createAgentSession, ModelRegistry, SessionManager } from "@juliusbrussee/caveman-code";

const { session } = await createAgentSession({
  sessionManager: SessionManager.inMemory(),
  authStorage: AuthStorage.create(),
  modelRegistry: ModelRegistry.create(AuthStorage.create()),
});

session.on("message", (msg) => console.log(msg.role, msg.text));
await session.prompt("Refactor src/auth.ts to use the new TokenStore.");

Talk to a running daemon over HTTP / WS via @juliusbrussee/caveman-sdk. API reference →

TypeScript monorepo, 9 packages — full layout in CLAUDE.md.

---

Acknowledgements

Caveman Code is a heavy fork of pi-code by Mario Zechner. We track upstream and contribute fixes back where generally useful.

From pi-code (upstream)	Caveman Code's own work
Agent runtime · MCP scaffolding · provider OAuth · repo map · slash-command parser · settings manager · skills loader · edit-format renderers · TUI components	Caveman Mode (4-layer compression) · goal loop · plan mode · cavemem integration · /tree session branching · architect/editor split · honest-bench harness

Also indebted to Aider (repo map + edit-format-per-model), Claude Code (settings/commands/skills/agents/.mcp.json formats — adopted verbatim, then extended), Codex (cave-as-MCP-server), RTK (optional bash-output compression layer), and Biome (single-binary lint/format).

Missing credit? Open an issue — we'll fix it fast.

---

License

MIT © Julius Brussee. Forked from pi-code (MIT © Mario Zechner).

Issues · Releases · Changelog · Docs

_{Caveman no waste token. Caveman ship.}