generated: '2026-09-19'
method: searched
source: https://api.berrergate.com/llms.txt
derived_from: openapi/berrergate-com-openapi.json
docs:
- https://api.berrergate.com/skill.md
- https://api.berrergate.com/v1/agent/utility/catalog
- https://berrergate.com/privacy/
base_url: https://api.berrergate.com
media_type: application/json
auth:
  style: >-
    None. No API key, token or account exists; every read and the free preview POSTs are anonymous. Paid
    routes are gated per request by x402 payment (HTTP 402 -> settle -> retry with PAYMENT-SIGNATURE), USDC
    on Base. See authentication/berrergate-com-authentication.yml.
  detail: authentication/berrergate-com-authentication.yml
idempotency:
  supported: true
  coverage: partial
  mechanism: caller-generated key in a request header, required (400 IDEMPOTENCY_KEY_REQUIRED without it)
  header: x-idempotency-key
  scope:
  - 'POST /v1/research'
  - 'POST /v1/agent/beta/query'
  - 'POST /v1/agent/trial/query'
  - 'POST /v1/agent/utility/inference'
  - 'POST /v1/agent/utility/spend-router'
  - 'POST /v1/agent/wisdom/provider-select'
  retention: undocumented
  description: >-
    Six of the nine POST routes declare an x-idempotency-key header parameter and the server enforces it —
    POST /v1/research without the header answered 400 {"ok":false,"error":"IDEMPOTENCY_KEY_REQUIRED"} on
    2026-09-19. Every route that can cost money carries it, which is the part that matters for an agent:
    the same key replays the same paid request rather than paying twice. The utility catalog states
    "provider_output_stored_for_idempotent_replay: true" and the privacy page that "ambiguous settlement
    states are reconciled rather than automatically retried" (auto_retry_after_ambiguous_settlement: false
    in the catalog). How long a key is honoured, and whether a key reused with a different body is rejected,
    are not published.
  gaps:
  - No key on POST /v1/agent/procurement/preview or POST /v1/agent/discovery/preview (free, cache-only reads over POST — low consequence).
  - No key on the A2A routes (POST /a2a/jsonrpc, POST /a2a/v1/message:send), which can trigger the same paid skills.
  - No retention window or conflict semantics documented.
dry_run_mode:
  supported: true
  status: documented
  mechanism: free preview routes plus the x402 quote step
  surfaces:
  - operation: 'POST /v1/agent/procurement/preview'
    cost: free
    description: 'A "free procurement preview" of coverage and decision context for a capability before paying (GET /v1/agent/procurement/catalog lists the ten capability keys, all free_preview: true).'
  - operation: 'GET /v1/agent/trial/catalog + POST /v1/agent/trial/query'
    cost: free
    description: 32 deterministic prewarmed "free-trial capability products" (not_for_sale true, payment_required false) that let an agent check whether BerrerGate already has intelligence for a capability "before paying".
  - operation: 'first POST without PAYMENT-SIGNATURE on any paid route'
    cost: free
    description: 'Per llms.txt: "Omit PAYMENT-SIGNATURE on the first POST to receive the canonical x402 payment requirement" — the 402 body states the exact amount, asset, network and payTo before anything is charged (observed on POST /v1/research: 20000 atomic USDC).'
  - operation: 'GET /v1/agent/utility/catalog'
    cost: free
    description: Publishes the dynamic price band (min/max atomic) and call caps for the inference product ahead of a call.
reversibility:
  grade: none
  docs: null
  note: >-
    No cancel, refund, void or reversal operation exists in the contract and no refund policy is published
    anywhere on the site (there is no Terms of Service page — /terms, /terms/ and /legal all 404). The
    write surface consists of one-shot paid deliveries (a decision packet, an evidence brief, an inference
    answer) settled by x402 with "one-shot settlement semantics"; the only post-purchase read is GET
    /v1/agent/beta/receipts/{receipt_id}. The provider's own mitigations are pre-commit — the free preview,
    the 402 quote, the required idempotency key and a hard "payment_before_provider_cost" ordering — not
    post-commit. Nothing below asserts a window the provider has not written down.
  write_surfaces:
  - operation: 'POST /v1/agent/beta/query'
    action: buy a $0.01 decision packet
    reversal: null
    window: null
    receipt: 'GET /v1/agent/beta/receipts/{receipt_id}'
  - operation: 'POST /v1/research'
    action: buy a $0.020 Evidence Brief
    reversal: null
    window: null
  - operation: 'POST /v1/agent/utility/inference'
    action: buy a bounded AI answer ($0.00125-$0.00225)
    reversal: null
    window: null
    note: The utility catalog states payment "is settled only after provider success and a positive direct-provider margin check" — a failed call is not charged, which is a pre-settlement guard, not a reversal.
  - operation: 'POST /v1/agent/utility/spend-router'
    action: paid read-only catalog route ($0.001)
    reversal: null
    window: null
    note: STAGED_DISABLED on 2026-09-19.
pagination:
  style: none
  note: Every list route (catalogs, sources, capabilities) returns the full set in one body; no cursor, page or limit parameters exist.
field_expansion: {supported: false}
sparse_fields: {supported: false}
metadata: {supported: false}
request_id:
  response_fields: [trace_id, request_id]
  header: null
  note: Both fields appear in every success and error body (identical UUIDs); no X-Request-Id header is returned.
versioning:
  style: URI prefix /v1
  detail: lifecycle/berrergate-com-lifecycle.yml
error_envelope:
  shape: '{"ok": false, "error": "<CODE>", "trace_id": "<uuid>", "request_id": "<uuid>"}'
  exception: 402 bodies are the x402 v2 PaymentRequirements object; JSON-RPC routes return {"jsonrpc":"2.0","error":{code,message,data}} with HTTP 200
  format: vendor-json (not RFC 9457)
  detail: errors/berrergate-com-problem-types.yml
rate_limit_signaling:
  status: 429
  headers: []
  published_caps: 'utility catalog: inference daily_max_calls 75, lifetime_max_calls 150, per_payer_daily_max_calls 25'
  detail: rate-limits/berrergate-com-rate-limits.yml
caching:
  observed: 'GET catalogs return cache-control: public, max-age=14400, s-maxage=300, stale-while-revalidate=60 with age and last-modified; paid POST errors return cache-control: no-store'
cors:
  access_control_allow_origin: '*'
  allow_methods: [GET, POST, OPTIONS]
  allow_headers: [content-type, x-bcg-client-version, x-bcg-admin-token, x-idempotency-key, x-agent-id, x-bcg-agent-id, x-bcg-telemetry-class, x-bcg-discovery-source, mcp-protocol-version, mcp-method, mcp-name, payment-signature]
  expose_headers: [payment-required, payment-response]
input_limits:
  - {operation: 'POST /v1/research', field: query, limit: '1-700 chars'}
  - {operation: 'POST /v1/research', field: max_evidence, limit: '1-5 (default 5)'}
  - {operation: 'POST /v1/agent/utility/inference', field: prompt, limit: '<=3200 (3,200 UTF-8 bytes per llms.txt)'}
  - {operation: 'POST /v1/agent/utility/inference', field: max_output_tokens, limit: '32-512'}
  - {operation: 'POST /v1/agent/discovery/preview', field: query, limit: '1-300 chars; country <=2; source_key <=80'}
  - {operation: 'POST /v1/agent/utility/spend-router', field: query, limit: '1-100 chars; country <=2; type <=40'}
privacy_conventions:
  raw_prompt_stored: false
  raw_agent_id_stored: false
  source: https://berrergate.com/privacy/ and every catalog body's privacy/telemetry block