ZeroBurn Documentation

Getting Started

ZeroBurn is a Python-based routing layer for OpenClaw and compatible AI agent frameworks. It intercepts model requests from your agent, evaluates each one against routing policies, and directs the request to the most cost-effective model capable of completing the task — whether that's a free local model running via Ollama, a less expensive cloud model, or a premium cloud model when one is genuinely needed.

The result: most users see significant reductions in monthly Anthropic and OpenAI spend with no perceived quality loss.

If you haven't purchased ZeroBurn yet, head back to the product page. If you have your license email, jump to Installation.

Installation

Requirements

• Python 3.10 or newer
• macOS (Apple Silicon or Intel) or Linux
• Ollama installed and running — get it from ollama.com
• At least one local model pulled (we recommend ollama pull llama3.1:8b to start)
• An OpenClaw installation or compatible Python-based AI agent framework

Install command

After purchase, your customer dashboard at app.ghosttai.com provides a one-line install command. It looks like:

curl -L "https://license.ghosttai.com/api/download/install?token=<your-token>" \
  -o /tmp/zeroburn.zip && openclaw plugins install /tmp/zeroburn.zip

The download token is unique to your purchase and expires after 24 hours. If yours has expired, click Regenerate on your dashboard for a fresh one.

Verify installation

After installing, run:

zeroburn status

You should see your edition (Community or Pro) and a confirmation that the license is active. If you see "Community" but purchased Pro, see Troubleshooting.

Configuration

ZeroBurn ships with sensible defaults — most users don't need to configure anything to start seeing savings. For tuning, configuration lives at ~/.zeroburn/config.yaml.

Common settings

• Local model: which Ollama model to route to (default: llama3.1:8b)
• Cloud fallback: which paid model to use when local isn't suitable
• Routing aggressiveness: from conservative (only obviously simple tasks go local) to aggressive (most tasks go local unless escalation rules trigger)
• Cache TTL: how long to cache identical requests before re-routing

Detailed configuration reference is in progress. For specific questions, contact [email protected].

Routing Policies

Routing decisions are made per-request based on signals from the prompt and context: estimated complexity, presence of code, length of context, whether the task has been seen before, and explicit policy rules.

Default policy

• Local first: routine queries, short responses, repeat patterns
• Cheap cloud: moderate complexity, longer context, when local model confidence is low
• Premium cloud: multi-step reasoning, code generation, anything that escalated from cheap cloud due to confidence checks

Custom rules

You can pin specific request patterns to specific models — for example, always route requests containing a specific keyword to a specific model, or always avoid sending code that matches a regex pattern to cloud providers. Details forthcoming as the custom rules system stabilizes.

Pro Features

Pro edition includes everything in Community plus:

• Multi-agent orchestration — coordinate two parallel agent instances for tasks that benefit from concurrency
• Nightly memory consolidation — automatically dedupes and compresses your context history for better routing decisions over time
• Daily spend cap (Cost Guard) — set a hard daily limit on cloud model spend; ZeroBurn refuses to escalate once the cap is hit
• Full analytics dashboard — see routing distribution, savings, and per-model costs over any time range
• Improvement proposals — periodic analysis of your usage to suggest config changes that would save more
• Priority email support

CLI Reference

ZeroBurn ships a CLI for inspection, configuration, and control. The core commands:

Troubleshooting

Install fails with "command not found: openclaw"

OpenClaw isn't installed or isn't on your PATH. Install OpenClaw first, then re-run the install command.

Install fails with "command not found: ollama"

Ollama isn't installed. Get it from ollama.com.

zeroburn status reports Community but I bought Pro

1. Make sure you installed the zip from your customer dashboard, not an older download
2. Check that the bundled license.json arrived inside the install package
3. Run zeroburn telemetry status to confirm the license server is reachable
4. If still stuck, email us with your purchase email

Lost or expired download link

Log in to your customer dashboard at app.ghosttai.com and click Regenerate on your install card. The new token is valid for another 24 hours.

Moving to a new machine

Community licenses are bound to one machine. To transfer, email us with your current zeroburn status output and we'll deactivate the old binding. Pro licenses have more flexibility — details in your customer dashboard.

FAQ