Agent System
LIVE
Six specialized AI agents orchestrated by AriBot. Each agent handles a specific domain — video editing, metadata, art generation, analytics, code, and library management. All inference runs locally via Ollama on the RTX A6000.
Six specialized AI agents orchestrated by AriBot. Each agent handles a specific domain — video editing, metadata, art generation, analytics, code, and library management. All inference runs locally via Ollama on the RTX A6000.
AriBot acts as the orchestrator — it receives natural language commands, routes them to the appropriate specialist agent, and aggregates results. Each agent has its own Goose recipe and model assignment optimized for its workload.
Heavy agents (CodeBot, EdBot) get the full 30B coder model for complex reasoning. Light agents (AnaBot, ArtBot, LibABot) run on the 14B model — fast enough for their workloads and leaves VRAM headroom for ComfyUI and Resolve.
| Agent | Model | VRAM | Warm Response |
|---|---|---|---|
| AriBot | qwen3-coder:30b-32k | ~22 GB | 2.3s |
| EdBot | qwen3-coder:30b-32k | ~22 GB | 2.3s |
| CodeBot | qwen3-coder:30b-32k | ~22 GB | 2.3s |
| AnaBot | qwen3:14b | ~9 GB | <1s |
| ArtBot | qwen3:14b | ~9 GB | <1s |
| LibABot | qwen3:14b | ~9 GB | <1s |
| Component | Role |
|---|---|
| Goose CLI v1.26.1 | Agent execution runtime with recipe-based task dispatch |
| Ollama | Local LLM inference server (port 11434) |
| FastAPI (port 8000) | Bridge server — SSE streaming from Goose to browser UI |
| 11 Goose Recipes | Specialized prompt templates per agent/task type |
| routing.yaml | Keyword-based task → agent → recipe routing table |
Sample frames from the footage library that EdBot, AnaBot, and LibABot analyze, score, catalogue, and prepare for rough cut assembly.
