OmniForge
Ask anything. Right agents. Right models.
One multimodal ask fans out across agents and tools. The waterfall proves which model ran each step.
Architecture
Self-contained monorepo: ingest → planner → parallel agents/MCP → Multi-LLM Brain → synthesizer → waterfall. No sibling runtime dependencies.
- Task-class routing: fast / structured / reasoning / vision
- Text, image, and voice transcript in one ask path
- In-repo FinOps budget + export gate