Connect the market. Then the model.

Start with the website

Import fourtyeightTech/metabrain into Vercel as a Next.js application. Keep the root directory at ., use npm ci to install and npm run build to build. Set METATRAY_MODE=live and METATRAY_FEED=rpc.

Add server variables RPC_HTTP_URL, CHAIN_ID, TOKEN_ADDRESS, MARKET_PROTOCOL=pons-v2 and PONS_FACTORY_ADDRESS. Use this project’s verified contracts. For a compatible V3 pool, choose MARKET_PROTOCOL=uniswap-v3 and supply V3_POOL_ADDRESS instead of the factory. Keep RPC credentials private; never prefix them with NEXT_PUBLIC_.

Redeploy after changing variables. No database or model download is needed for this recent-swap view. Missing settings appear in the feed console. Optional BLOCK_EXPLORER_URL adds transaction links. Synthetic previews require an explicit METATRAY_MODE=demo.

Add persistent history and model jobs

Provision Postgres and run the supplied migration. Configure the actual chain ID, token, start block and verified protocol addresses. Run the Node indexer on a persistent host. The repository’s environment template and doctor command describe the required inputs.

Verify a real swap against its transaction receipt. Switch the website to METATRAY_FEED=indexed and supply a limited DATABASE_READ_URL. Keep worker credentials private. This enables the persistent paper ledger and optional GPU job queue; the direct RPC observer has neither.

Prepare the Meta worker

Establish the rights basis for the intended model use, obtain access to the required feature models and select immutable checkpoint revisions. Prepare the model volume and start the supplied Python service on a suitable GPU host.

The integration imports TribeModel, constructs events from stimulus.mp4, and calls the upstream prediction interface. Run the operator smoke test and measure real processing latency before treating the connection as ready.

Vercel does not host this persistent GPU workload. Setting a website environment variable alone cannot produce a cortical prediction.

Check the complete connection

1. Confirm the observed token and quote asset are correct.
2. Match an event to its source transaction and block.
3. Inspect the generated audiovisual input and its cutoff.
4. Verify a completed model result, surface dimensions and provenance.
5. Check freshness labels, outage behavior and the paper rule.
6. Review the public terms, private contact route and actual data practices before launch.

Deployment references