Run models

Runs Leni / Realsage models from your prompt, optional attachments, and session context. Supports JSON and multipart as described in File uploads.

POST/users/run-models

Sandbox run-models: rate limit + sandbox quota apply; no credit balance check Development / production run-models: rate limit + credit gate apply; sandbox quota is not used.

Request fields

Field	Type	Required	Description
question	string	Required	The prompt you want the model to answer. Must be non-empty after leading and trailing whitespace is removed.
files	array	Optional	Optional file attachments. In JSON, send an array of objects using the base64 shape described in File uploads. In multipart requests, add one or more parts named files.
filesBase64	array	Optional	Optional JSON-only alias for files. Uses the same base64 object shape. When filesBase64 is present on the body, it is used instead of files so you can reserve files for multipart-only uploads.
sessionId	string	Optional Required for session history	You must send sessionId on every follow-up request if you want the question to be part of that session's history. Reuse the same id from a previous response so each turn is recorded in one thread and appears when you list or fetch messages for that session. When you omit sessionId with a valid API key, the service may create a new session and return its id—save that id and include it on your next request to continue the conversation.
tone	string	Optional	How the assistant should sound. One of: professional, friendly, authoritative, consultative, funny. Defaults to professional.
fastMode	boolean	Optional	When true, requests a faster response path where the upstream model supports it. Defaults to false.
uploadedAttachments	array	Optional	Metadata for attachments already stored (for example S3 references). This is not where you send raw file bytes—use files or filesBase64 for that.
userMemoryAttachmentsIds	string[]	Optional	Ids of user-memory attachments to include. If you omit this array, the service may derive ids from taggedMemoryIds.
orgMemoryAttachmentsIds	string[]	Optional	Ids of organisation-memory attachments to include. If you omit this array, the service may derive ids from taggedMemoryIds.
taggedMemoryIds	array	Optional	Memory items referenced in the prompt (for example [{ memoryType, id }]). When the explicit attachment id arrays are omitted, the service can derive them from this list.
customAnalyst	object	Optional	Use a custom analyst you created instead of a built-in model. Supply the analyst id and display fields; the service maps this to a custom-model request. After creating an analyst, set modelType to custom-model in user or send this object—see Custom analysts.
wait	boolean \| string	Optional	Controls how long the gateway waits before responding. Set true or the string "true" (case-insensitive) to disable the default ~5 second HTTP deadline and wait for the full Leni response. Omit or use false to allow an early response with messageId, sessionId, and polling hints while the model may continue in the background.

Fields inside user

After merging with your API key, the request typically carries:

Required: id, organisationId, modelType.
Optional: modelId, buildingIds, leni_report_builder.
Custom analysts: use customAnalyst or configure custom-model per Custom analysts.

Optional files

Multipart: append one or more parts named files.
JSON: include a files or filesBase64 array using the shared base64 shape (see File uploads). Decoded files are sent as files parts with the rest of the payload.
Tagged memory: if you send taggedMemoryIds, the backend can derive userMemoryAttachmentsIds and orgMemoryAttachmentsIds when those arrays are not supplied explicitly.

Example: JSON with base64 file

JSON

{
  "question": "Summarize the attached file in two sentences.",
  "user": {
    "id": "00000000-0000-0000-0000-000000000000",
    "organisationId": "00000000-0000-0000-0000-000000000000",
    "modelType": "leniq-pro"
  },
  "tone": "professional",
  "files": [
    {
      "filename": "note.txt",
      "contentType": "text/plain",
      "contentBase64": "SGVsbG8gd29ybGQ="
    }
  ]
}

Example: multipart (form-data)

Key	Type	Value
question	Text	What is in this PDF?
user	Text	{"modelType":"leniq-pro"} (add ids as needed)
files	File	Select local file(s)

Response deadline and polling

By default, the gateway may apply a short HTTP response deadline (about five seconds) while Leni /run-models is still running. If the upstream call has not finished in time, you can receive an early JSON payload that still identifies the persisted row in chat.message, for example messageId, sessionId, responseDeadlineReached: true, and a hint on how to poll. The model request may continue in the background.

Send wait as true (boolean or string "true") on the inbound JSON or multipart text field to disable that deadline and wait for the full Leni response. Poll with GET /users/me/chat-messages/:messageId or GET /users/me/chat-sessions/:sessionId/messages when you received a deadline response.

Response

Response JSON depends on the model and options selected. Inspect the status code and body for messages, session identifiers, and tool output.

Code examples

const res = await fetch(`${process.env.API_BASE}/users/run-models`, {
  method: "POST",
  headers: {
    "Content-Type": "application/json",
    "X-Api-Key": process.env.LENI_API_KEY!,
  },
  body: JSON.stringify({
    question: "What is in this PDF?",
    user: { modelType: "leniq-pro" },
    tone: "professional",
  }),
});

const data = await res.json();

Try it

Using key: leni_sk_sandbox_a3f8••••

{
  "question": "Summarize the attached file in two sentences.",
  "user": {
    "id": "00000000-0000-0000-0000-000000000000",
    "organisationId": "00000000-0000-0000-0000-000000000000",
    "modelType": "leniq-pro"
  },
  "tone": "professional"
}