Everything that touches The Graph lives here, kept apart from the marketplace packages so it can be opened as a standalone repository (a hackathon submission needs a public repo) without dragging the rest of Ainize with it.
What belongs here:
- Subgraph → dataset: pipelines that turn a Standardized Subgraph query (Messari-style schemas,
ERC-4626 tokenized vaults, Agent0/ERC-8004) or a Substreams package into a canonical Ainize teaching
dataset (
{prompt, answer}rows) with provenance — which subgraph, which query, which block, row hashes. - Substreams modules we author or extend, and any standardized-schema work.
- The MCP client side that calls The Graph's Subgraph MCP and hands its rows to
packages/mcp. - The benchmark harness that measures the quality difference between: base model · base + Subgraph MCP · base + Ainize knowledge · both — accuracy, latency, tokens, cost, hallucinated-address rate, side effects.
- Demo scripts, evidence and the submission write-up.
What does not belong here: the marketplace itself (packages/*), the trainer (/mnt/newdata/qwen3.8),
or anything that would break if The Graph were removed.
Live data only: every pipeline reads from a Graph provider (Subgraph Studio API key, The Graph Market for Substreams). Mocked, local-only or static datasets do not qualify for the tracks this targets and are not accepted here either.