# ERPClaw by AvanSaber

- **Event:** [Meta Global AI Developer Hackathon](https://cerebralvalley.ai/e/meta-ai-developer-hackathon)
- **When:** Oct 26, 12:00 PM – Nov 4, 9:00 PM (EST)
- **Where:** Online
- **Team:** [Nikhil J](https://cerebralvalley.ai/u/jathar)
- **GitHub:** https://github.com/avansaber/erpclaw/tree/v3
- **Demo video:** https://github.com/avansaber/erpclaw/releases/download/v3-preview-2026-11-04/ERPClaw-v3-3-minute-demo.mp4
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5303

Public repository: https://github.com/avansaber/erpclaw
ERPClaw v3 source branch: https://github.com/avansaber/erpclaw/tree/v3

ERPClaw v3 is a from-scratch, GPL-3.0-only ERP engine that lets small-business owners work through conversation. Its command-line interface makes the business workflow directly accessible: ask for an invoice, answer any clarification, inspect the engine's exact approval card, and explicitly approve or decline the proposed change.

The engine resolves business records and calculates the figures. The model proposes an operation; the owner controls approval. Before committing, the engine rechecks the approved card. Money uses exact decimal arithmetic, and the books use PostgreSQL.

Our hackathon entry is the full ERPClaw v3 engine, CLI and included first-party plugins. The video demonstrates one invoice workflow. The v3 engine, ledger, approval cards, CLI and included first-party plugins were developed from scratch during 27 October to 4 November 2026 for this event. The contribution record identifies that development period and Meta's specific role. The CLI renders the card's line items, proposed book effects and warnings, with approval defaulting to No. The recorded flow creates and submits an invoice through one explicit approval card.

The aim is practical: let owners describe their work naturally while keeping business rules and financial authority in software they can inspect and self-host. Where a capability is unavailable, ERPClaw is designed to say so rather than invent an answer.

During development, Meta Model API's Muse Spark 1.3 Contributor produced a focused stock-plugin helper and reviewed a separate engine component. Both components are included in the prepared submission branch. The warehouse helper is called by stock setup. The development candidate has 160 recorded passing pure unit cases. The recorded CLI conversation uses configured Qwen 3.5 9B through OpenRouter; Meta's contribution is to engine development.

## More from Meta Global AI Developer Hackathon

- [Deskverse](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5300)
- [KushZombie](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5301)
- [OODA-Aegis](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5302)
- [Ride AI](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5304)
- [Easy Onboarding](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5305)
- [OmniOps](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5306)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5303. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt