🔌Nexus API
The Nexus API is a hosted, read-only HTTP boundary over indexed Nexus state.
The architecture figure separates the hosted service from your application: the provider owns indexing and projections, while your code owns presentation, cursor persistence, retry policy, and secret custody. The API is a read surface; state changes still use a wallet-approved CLI or SDK transaction flow.
The sequence keeps credentials server-side and treats SSE as an invalidation and replay signal rather than a substitute for authoritative REST reads. A disconnect, replay gap, rate limit, stale projection, or envelope-free timeout becomes an explicit degraded state followed by the documented retry or re-read action.
Start here
Tutorial — configure a provider-issued key, read resources, stream events, assemble a raw-HTTP dApp, and port it to React
Connect a dApp — keep the key in a backend relay and expose only same-origin public data to a browser
Raw REST and SSE consumer — the typed-client page now uses page-local public types and standard fetch/EventSource-compatible behavior
API reference — scopes, resources, envelopes, status values, events, rate limits, and errors
Troubleshooting — diagnose credentials, projection lag, cursors, rate limits, and service responses
Configure the hosted consumer boundary
Obtain NEXUS_API_KEY from the provider through the agreed secure out-of-band channel. The public consumer contract does not assume a self-service key dashboard, a public credential-minting service, or a live service at the placeholder URL. Store the key in a secret manager or a mode-0600 temporary file, send it only from your server-side relay, and remove it after a short-lived test.
What it serves
/executions, walks, events, verdicts, failures, payment ledger
DAGExecution traces and payment projections
/payments, /priority-fee-vault/events
Execution payments and fee-vault activity
The public projection may omit Task rejection or Invocation fields. If a deployment does not publish a field or event, use the provider’s documented schema and do not infer the missing state from an empty response.
Read, react, and recover
REST responses are the source for the view; the SSE stream is a timely, replayable signal that tells the application when to re-read. List pages use an opaque next_token, statuses are open sets, and a missing row can mean projection lag rather than absence. The tutorial and document show the exact envelopes and cursor rules.
Last updated

