generated: '2026-08-09'
method: searched
source: https://legal.ge/llms.txt + openapi/legal-ge-public-apis-openapi.yml
summary: >-
  Cross-cutting request/response semantics for the legal.ge Public API, captured from the published
  llms.txt agent contract and the OpenAPI 3.1 spec. This is a small, keyless, strictly read-only
  API: no authentication, no pagination cursors, no idempotency contract (there is nothing to make
  idempotent — every operation is a GET or a side-effect-free POST search).
authentication:
  style: none
  keyless: true
  notes: >-
    No securitySchemes are declared in the OpenAPI and none are documented. Access is gated only by
    per-IP rate limits. Sending an inquiry to a specialist requires an interactive sign-in on
    legal.ge and is intentionally not exposed through the API.
  see: authentication/legal-ge-public-apis-authentication.yml
idempotency:
  documented: false
  applicable: false
  notes: >-
    No Idempotency-Key header or equivalent is documented, and none is needed: the API is read-only.
    POST /api/ask is a search that creates no resource, so retrying it is inherently safe.
pagination:
  style: none
  notes: >-
    /api/ask accepts a `limit` (1-50, default 20) and returns meta.total / meta.returned, but there
    is no cursor, offset or page parameter — a caller cannot page past the limit.
    /api/specialists/search returns up to 20 results with no paging parameter at all.
  params: [limit]
  response_fields: [meta.total, meta.returned, meta.fallback_used, meta.took_ms]
localization:
  param: locale
  values: [ka, en, ru]
  spec_default: ka
  mcp_default: en
  notes: >-
    Trilingual by design — keywords, services and specialist bios are translated. Profile URLs
    returned in `profile_url` are canonical and locale-prefixed (/ka/, /en/, /ru/).
field_selection:
  supported: false
metadata:
  supported: false
request_tracing:
  request_id_header: null
  documented: false
versioning:
  scheme: none-in-path
  current: 1.0.0
  notes: >-
    The API is unversioned on the wire — no /v1 prefix and no version header. The only version
    signal is info.version in the OpenAPI document.
  see: lifecycle/legal-ge-public-apis-lifecycle.yml
error_envelope:
  shape: custom
  rfc9457: false
  media_type: application/json
  example: '{"error": {"code": "QUERY_TOO_LONG", "message": "query exceeds 500 characters", "field": "query", "constraint": {"max_length": 500}, "received_length": 712}}'
  fields:
    - {name: error.code, description: 'Machine-readable, stable across versions.'}
    - {name: error.message, description: 'Human-readable, safe to surface to end users.'}
    - {name: error.field, description: 'Offending request field, when field-scoped.'}
    - {name: error.constraint, description: 'Machine-readable constraint that failed; shape depends on code.'}
  see: errors/legal-ge-public-apis-problem-types.yml
rate_limit_signaling:
  status: 429
  headers: [Retry-After]
  body_field: error.constraint.retry_after_seconds
  notes: >-
    429 responses carry both a Retry-After header (seconds) and a structured payload with
    error.code = "RATE_LIMITED". No X-RateLimit-Limit / -Remaining / -Reset headers are documented.
  see: rate-limits/legal-ge-public-apis-rate-limits.yml
agent_conventions:
  citation_fields: [full_name, profile_url, company.name, matched_categories, professional_orgs]
  ranking:
    exposed: false
    surfaceable_field: ranking_note
    notes: >-
      Ordering blends category match strength with internal quality and verification signals; exact
      inputs are intentionally not exposed. The `ranking_note` string is safe to show end users.
  trust_contract:
    - Every specialist returned has verification_status='verified'.
    - contact.email and contact.phone are populated only when the specialist opted in; otherwise null.
    - Anonymous inquiry submission is not possible — agents must return profile URLs for click-through.
  robots_policy: >-
    robots.txt names eleven AI crawler user-agents (GPTBot, ChatGPT-User, OAI-SearchBot,
    Google-Extended, Claude-Web, anthropic-ai, PerplexityBot, Applebot-Extended,
    Meta-ExternalAgent, Amazonbot, CCBot) and Allows them on the public API paths while
    Disallowing the rest of /api/*. Bytespider is fully disallowed.
cross_links:
  authentication: authentication/legal-ge-public-apis-authentication.yml
  errors: errors/legal-ge-public-apis-problem-types.yml
  lifecycle: lifecycle/legal-ge-public-apis-lifecycle.yml
  rate_limits: rate-limits/legal-ge-public-apis-rate-limits.yml
  mcp: mcp/legal-ge-public-apis-mcp.yml