Gradio UI to build and search TileDB vector indexes, run RAG-style evaluations on golden datasets, and compare indexes. Uses LangGraph and Anthropic (Claude) for the agent path.
- Python 3.11+
ANTHROPIC_API_KEY— required for the RAG agent and optional LLM judge features (standard Anthropic env var).
From the repo root (uses uv):
uv syncOptional dev tools: uv sync --extra dev.
uv run rag-evaluator-uiOpens the app at http://127.0.0.1:7860 by default. Override bind address and port:
uv run rag-evaluator-ui --host 0.0.0.0 --port 7860Agent instructions for this repo live under .agents/skills/:
| Skill | Path | Use when |
|---|---|---|
| Weights & Biases (W&B / Weave) | .agents/skills/wandb-primary/SKILL.md |
Runs, metrics, artifacts, Weave traces, evals, scorers |
| TileDB Vector Search CLI | .agents/skills/tiledb-vector-search-cli/SKILL.md |
tiledb vs — build indexes from docs, semantic search, CLI workflows |
Attach or register these with your agent tooling if you want assistants to load them for W&B or tiledb vs work.
