generated: '2026-09-18'
method: searched
source: >-
  https://docs.neuralverge.ai/authentication, /errors-and-rate-limits,
  /ai-research, /search, /mcp-server, plus derivation from
  openapi/neuralverge-api-openapi.json.
authentication:
  style: bearer-token
  header: 'Authorization: Bearer <API_KEY>'
  note: Same token used for REST and the remote MCP server.
base_url: https://api.neuralverge.ai
path_convention:
  pattern: /functions/v1/{action}
  note: Supabase Edge Functions; each capability is an action verb (run-*, get-*).
execution_model:
  synchronous:
    note: >-
      Most endpoints return their result in the same response (search, extract,
      all data-source lookups, run-agent).
  asynchronous:
    operations: ['POST /functions/v1/run-research']
    pattern: >-
      run-research returns a session_id immediately; poll GET
      /functions/v1/get-session-status every 2-5s until status is complete or
      failed.
pagination:
  style: page-number
  note: >-
    LinkedIn search/employee endpoints use startPage plus maxResults/maxItems;
    run-amazon-product-search paginates result cards. Search uses max_results as
    a count limit. No cursor-based pagination and no Link headers.
  params: [startPage, maxResults, maxItems, max_results]
idempotency:
  supported: false
  coverage: none
  note: >-
    No Idempotency-Key header or equivalent replay-protection mechanism is
    documented on any write/action endpoint. Every run-* call is a metered POST
    that spends points; a retry re-executes and re-charges. Agents must guard
    against duplicate submission client-side.
reversibility:
  state: na
  note: >-
    The action surface is metered lookups and research/extraction jobs that
    return data; none creates a durable, caller-mutable resource, so there is no
    cancel/refund/void/undo operation and nothing to reverse. Points spent are
    consumed by the run (402 on exhaustion) and are not refundable via the API.
    Reversibility, dry-run and idempotency are all na/none for this reason.
dry_run_mode:
  supported: false
  state: na
request_tracing:
  request_id_header: null
  note: No request-id / trace-id response header is documented.
versioning:
  scheme: uri-path
  current: v1
  note: Path segment /functions/v1/; OpenAPI info.version 1.0.0.
error_envelope:
  shape: '{"error": "<message>"}'
  rfc9457: false
  ref: errors/neuralverge-api-problem-types.yml
rate_limit_signaling:
  headers: none
  note: >-
    No X-RateLimit-*/RateLimit-*/Retry-After headers documented. Limits are
    enforced by server-side queueing; a request that cannot start within 60s
    fails with 429. See rate-limits/neuralverge-api-rate-limits.yml.
  ref: rate-limits/neuralverge-api-rate-limits.yml