Caveman Mode

Cut ~79% of tokens from your AI context without losing functionality.

Why Caveman Mode?

AI agents consume tokens for every file they read. In large projects, context windows fill up fast. Caveman Mode provides stripped-down templates that preserve structure but eliminate:

• Comments and explanations
• Examples and sample data
• Verbose headers
• Decorative whitespace

Token Budget

File	Standard	Caveman	Savings
AGENTS.md	~500 tokens	~120 tokens	76%
MEMORY.md	~400 tokens	~80 tokens	80%
TASKS.md	~300 tokens	~60 tokens	80%
RUNBOOK.md	~400 tokens	~80 tokens	80%
Total	~1,600	~340	~79%

Fresh Install

Via npx (Recommended):

npx multimodel-dev-os@latest init --caveman

Via shell script installer:

curl -fsSL https://raw.githubusercontent.com/rizvee/multimodel-dev-os/main/scripts/install.sh | bash -s -- --caveman

Convert Existing Project

Replace root files with caveman variants:

cp .ai/templates/AGENTS.caveman.md AGENTS.md
cp .ai/templates/MEMORY.caveman.md MEMORY.md
cp .ai/templates/TASKS.caveman.md TASKS.md
cp .ai/templates/RUNBOOK.caveman.md RUNBOOK.md

Update config:

# .ai/config.yaml
mode: "caveman"

Switch Back to Standard

# Re-download standard templates
curl -fsSL .../AGENTS.md -o AGENTS.md
# Or restore from git
git checkout -- AGENTS.md MEMORY.md TASKS.md RUNBOOK.md

Caveman Writing Rules

If you're writing your own caveman-style files:

1. No comments — every token must be actionable
2. No examples — agents know common patterns
3. No headers longer than 3 words
4. No blank lines between sections
5. Use abbreviations: dev, deps, cfg, env
6. Use null for missing values
7. Tables over prose — always

When NOT to Use Caveman Mode

• New projects where you need detailed templates as guides
• Teams where multiple humans need to read the files
• When you have plenty of context window budget
• When onboarding new team members who need explanations