Firebat is built as a 3-LLM collaborative project — designed, written, and refined together by humans and three frontier AI models. Each model contributed at different stages and in different roles.
Anthropic — Claude
- Bulk of the recent codebase (CMS V2 builder, memory system, AI tool framework, Function Calling integration, hexagonal refactor)
- Code review and architectural decisions
- Used via Claude Code (CLI) and the Anthropic API
- Co-author of the FIREBAT_BIBLE (architectural principles, ports & adapters design, single-tenant constraints)
- Coding contributions via Codex CLI
- Used via Codex CLI, GPT-5 / GPT-5.5 series
Google — Gemini / Gemini CLI
- Initial Firebat prototype — early version of the codebase that became the foundation for the current architecture
- Co-author of the FIREBAT_BIBLE (alongside GPT)
- Coding contributions via Gemini CLI
- Used via Gemini CLI, Gemini 2.5 / 3 series (Pro / Flash) and Vertex AI
Firebat itself is an AI-powered Visual Automation Agent (VAA) — a platform where multiple LLMs orchestrate code, content, and automation. It is fitting that the platform itself was built by the same kind of AI ↔ human collaboration it now enables.
The three frontier models each bring different strengths: Claude excels at large refactors and architectural reasoning, GPT excels at thorough specification writing and edge-case analysis, Gemini excels at fast prototyping and multi-modal reasoning. This project leans into all three.
This project is eligible for the following OSS support programs:
- Anthropic Claude for OSS — Claude Pro/Max free tier
- OpenAI Codex for OSS — Codex CLI free tier
- Google AI Pro for OSS — Gemini CLI free tier
License: AGPL-3.0 — dual licensing available for commercial closed-source deployment (contact maintainer).