generated: '2026-07-19'
method: searched
source: https://docs.lang.ai/
summary: >-
  Lang.ai exposes a small, flat REST surface over a per-tenant subdomain. Everything is
  JSON in and JSON out (except project creation, which is a multipart/related upload),
  authentication is a single bearer API token, and the API is versioned in the URI path.
  There is no pagination, field expansion, sparse-fieldset or request-id contract
  documented; idempotency is offered on document writes through a client-supplied
  document id.
base_url:
  template: https://{company}.lang.ai/api/v1
  note: >-
    Per-tenant deployment — the subdomain is the customer's own Lang.ai instance.
  documented: https://company.lang.ai/api/v1
authentication:
  style: bearer-token
  header: Authorization
  format: 'Bearer <api-token>'
  issued_from: Settings section of your Lang.ai instance
  rotation: >-
    Generating a new token invalidates the previously issued ones — rotation is
    single-active-token, so a rotation is a hard cutover.
  applies_to: every request
  artifact: authentication/langai-authentication.yml
  docs: https://docs.lang.ai/#authentication
idempotency:
  supported: true
  mechanism: client-supplied-resource-id
  header: null
  parameter: id
  operations:
    - saveDocument
  semantics: >-
    On POST /documents the optional "id" field is the ID of the document to save; if a
    document with the same ID already exists it will be overwritten. Re-sending the same
    document with the same id is therefore an upsert rather than a duplicate insert,
    which makes retries of saveDocument safe.
  scope: per project document store
  retention: not documented
  note: >-
    This is client-supplied-id upsert idempotency, NOT an Idempotency-Key request header
    (RFC-style). Lang.ai documents no idempotency key header, and no idempotency contract
    for createProject, which is not safe to retry blindly.
  docs: https://docs.lang.ai/#documents
pagination:
  supported: false
  note: >-
    GET /projects returns the full list of projects as a bare JSON array with no
    pagination parameters, cursor, or envelope documented.
field_expansion:
  supported: false
sparse_fieldsets:
  supported: false
metadata:
  supported: true
  operations:
    - saveDocument
    - createProject
  shape: >-
    saveDocument accepts a free-form "metadata" object of key/value pairs on the
    document. New metadata values passed via the API become available to use in the
    project's setup section and can be used in the project's Dashboard. createProject
    accepts extra CSV columns in the uploaded dataset which are used as metadata.
  docs: https://docs.lang.ai/#documents
request_tracing:
  request_id_header: not documented
versioning:
  scheme: uri-path
  current: v1
  artifact: lifecycle/langai-lifecycle.yml
error_envelope:
  format: json
  rfc9457: false
  note: >-
    JSON is returned by all API responses, including errors. Lang.ai documents the HTTP
    status codes it returns (400/401/404/429/500) but publishes no field-level error
    body schema and does not use application/problem+json.
  artifact: errors/langai-problem-types.yml
  docs: https://docs.lang.ai/#errors
rate_limiting:
  supported: true
  documented_limits: false
  signal_headers: not documented
  status_code: 429
  note: >-
    "To avoid abuse and ensure that the system remains online for every user, every API
    request will need to provide valid credentials and the requests can be rate limited
    if they exceed the limitations." No numeric limits, quota windows or RateLimit
    response headers are published.
  docs: https://docs.lang.ai/#rate-limit
content_types:
  request:
    - application/json
    - multipart/related
    - text/csv
    - text/plain
  response:
    - application/json
dates:
  format: ISO 8601
  note: >-
    The saveDocument "date" field supports valid ISO 8601 dates and defaults to the
    request time when omitted.