The gateway checks the app-scoped credential, declaration, install grant, tenant policy, quotas and input schema. Existing grants are not expanded on publish. Public self-install auto-approval, when enabled, can grant required and optional declarations together; the Store has no per-capability checkbox. Use the SDK capabilities/requestCapability methods for missing access. User-scoped Drive and Calendar calls need X-Manaurum-User-Context. Directory, people, Telegram and other sensitive surfaces also need approval. The generated reference below lists each registered method, context and schemas directly from backend definitions.
Without an explicit provider/model, os.ai.complete uses the app’s selected backend, workspace default, then an available managed model. Managed calls share Assistant spending limits; BYOK is metered separately. AI Off or revoked access blocks calls. Forward user context for interactive calls; background calls may select X-Manaurum-Workspace-Id, otherwise exactly one eligible workspace is required. Read os.ai.providers for effective completion backend, platform_fallback and provider health. Embeddings and OCR require configured integrations; speech uses the selected model’s provider with a matching tenant or platform voice key. Images use BYOK and a tenant flag. Email/SMS use Resend/Twilio integrations. Missing configuration returns 412; shared spending caps return 429. log_prompt: false keeps supported AI input/output out of the call log without removing spend accounting.
os.kv persists app key/value data; os.files supplies presigned upload/download and list/delete. os.drive uses a two-step stage/publish flow or explicitly granted user folders; delete moves to Trash. os.calendar reads/creates user events. os.locations is read-only (list capped at 500); os.people includes removed invitees for 30 days. os.telegram posts via a workspace bot, with 20 sends/edits per chat per minute and callbacks to the app. os.apps.call, os.events.emit and os.apps.bulk_export implement RPC, async events and streaming datasets. os.secrets, os.tenant_config and os.compliance.audit_query expose scoped configuration/audit. Daily quotas return 429 quota_exceeded. Handle documented errors and cancellation rather than promising universal availability.
API reference
Capability
Context
Source contracts
os.ai.completev1
app
Input schema
json
{
"$schema": "https://json-schema.org/draft/2020-12/schema",
"type": "object",
"required": [
"messages"
],
"additionalProperties": false,
"properties": {
"provider": {
"type": "string",
"enum": [
"anthropic",
"deepseek",
"gemini",
"groq",
"openai"
],
"description": "Optional. When omitted, the OS resolves the tenant's configured provider automatically. The chosen provider is echoed in the response."
},
"model": {
"type": "string",
"minLength": 1,
"maxLength": 256,
"description": "Optional. When omitted, a sensible default for the resolved provider is used. Echoed in the response."
},
"messages": {
"type": "array",
"minItems": 1,
"items": {
"type": "object",
"required": [
"role",
"content"
],
"additionalProperties": false,
"properties": {
"role": {
"type": "string",
"enum": [
"system",
"user",
"assistant"
]
},
"content": {
"description": "A string, or a list of parts: {type: 'text', text} and, in user messages, {type: 'image', file_key}. At most 8 images per call, 20 MB together; the resolved model must read images.",
"oneOf": [
{
"type": "string"
},
{
"type": "array",
"minItems": 1,
"items": {
"oneOf": [
{
"type": "object",
"required": [
"type",
"text"
],
"additionalProperties": false,
"properties": {
"type": {
"const": "text"
},
"text": {
"type": "string"
}
}
},
{
"type": "object",
"required": [
"type",
"file_key"
],
"additionalProperties": false,
"properties": {
"type": {
"const": "image"
},
"file_key": {
"type": "string",
"minLength": 1,
"maxLength": 1024,
"pattern": "^(?!/)(?![\\s\\S]*\\\\)(?!(?:[\\s\\S]*/)?\\.\\.(?:/|$))",
"description": "Key of a JPEG, PNG or WebP image this app stored with os.files.upload, at most 5 MB."
}
}
}
]
},
"contains": {
"type": "object",
"required": [
"type",
"file_key"
],
"additionalProperties": false,
"properties": {
"type": {
"const": "image"
},
"file_key": {
"type": "string",
"minLength": 1,
"maxLength": 1024,
"pattern": "^(?!/)(?![\\s\\S]*\\\\)(?!(?:[\\s\\S]*/)?\\.\\.(?:/|$))",
"description": "Key of a JPEG, PNG or WebP image this app stored with os.files.upload, at most 5 MB."
}
}
},
"minContains": 0,
"maxContains": 8
}
]
}
},
"if": {
"properties": {
"role": {
"enum": [
"system",
"assistant"
]
}
}
},
"then": {
"properties": {
"content": {
"oneOf": [
{
"type": "string"
},
{
"type": "array",
"items": {
"type": "object",
"required": [
"type",
"text"
],
"additionalProperties": false,
"properties": {
"type": {
"const": "text"
},
"text": {
"type": "string"
}
}
}
}
]
}
}
}
}
},
"temperature": {
"type": "number",
"minimum": 0,
"maximum": 2
},
"max_tokens": {
"type": "integer",
"minimum": 1,
"maximum": 200000
},
"log_prompt": {
"type": "boolean",
"default": true,
"description": "Optional, default true. Pass false when the text carries the end user's private content: the platform then records the call's tokens, cost and attribution exactly as usual but stores a length + digest placeholder in place of the prompt, the system prompt and the model's answer."
}
}
}
os.ai.embedv1
app
Input schema
json
{
"$schema": "https://json-schema.org/draft/2020-12/schema",
"type": "object",
"required": [
"provider",
"model",
"input"
],
"additionalProperties": false,
"properties": {
"provider": {
"type": "string",
"enum": [
"gemini",
"openai"
]
},
"model": {
"type": "string",
"minLength": 1,
"maxLength": 256
},
"input": {
"oneOf": [
{
"type": "string",
"minLength": 1
},
{
"type": "array",
"minItems": 1,
"items": {
"type": "string",
"minLength": 1
}
}
]
},
"log_prompt": {
"type": "boolean",
"default": true,
"description": "Optional, default true. Pass false when the text carries the end user's private content: the platform then records the call's tokens, cost and attribution exactly as usual but stores a length + digest placeholder in place of the prompt, the system prompt and the model's answer."
}
}
}
{
"$schema": "https://json-schema.org/draft/2020-12/schema",
"type": "object",
"required": [
"text"
],
"additionalProperties": false,
"properties": {
"text": {
"type": "string",
"minLength": 1,
"maxLength": 20000,
"description": "What to say, as plain text or Markdown: code blocks are spoken as a short placeholder, links as their label, and formatting marks are dropped. Up to 4,000 characters are spoken; longer text is cut at a sentence end and the response says truncated: true."
},
"model": {
"type": "string",
"enum": [
"gemini-3.8-flash-lite-tts",
"gemini-3.8-flash-tts",
"gpt-4o-mini-tts"
],
"description": "Optional speech model. Without it the app speaks with the model a workspace admin set for it in Settings, else Manaurum's default."
},
"voice": {
"type": "string",
"enum": [
"Achernar",
"Achird",
"Algenib",
"Algieba",
"Alnilam",
"Aoede",
"Autonoe",
"Callirrhoe",
"Charon",
"Despina",
"Enceladus",
"Erinome",
"Fenrir",
"Gacrux",
"Iapetus",
"Kore",
"Laomedeia",
"Leda",
"Orus",
"Puck",
"Pulcherrima",
"Rasalgethi",
"Sadachbia",
"Sadaltager",
"Schedar",
"Sulafat",
"Umbriel",
"Vindemiatrix",
"Zephyr",
"Zubenelgenubi",
"alloy",
"coral",
"en-us-bodi",
"nova",
"onyx",
"sage"
],
"description": "Optional voice of the model that speaks (default: that model's default voice, or the voice set for the app). With an explicit model, a voice of another model's provider is 400 unknown_voice; without one, it gives way to the default voice."
},
"lang": {
"type": "string",
"minLength": 2,
"maxLength": 16,
"description": "Optional ISO-639-1 language of the text (e.g. 'ru'): the code-block placeholder is spoken in it."
}
}
}
os.ai.transcribev1
app
Input schema
json
{
"$schema": "https://json-schema.org/draft/2020-12/schema",
"type": "object",
"required": [
"audio_base64"
],
"additionalProperties": false,
"properties": {
"audio_base64": {
"type": "string",
"minLength": 1,
"maxLength": 35000000,
"description": "Base64-encoded audio (standard alphabet). Max 25 MB decoded — the upstream provider's upload limit."
},
"mime_type": {
"type": "string",
"minLength": 1,
"maxLength": 100,
"description": "Container MIME type of the audio (default audio/webm). Browsers differ: Chrome MediaRecorder emits webm/opus, iOS Safari emits audio/mp4 — pass what you recorded."
},
"model": {
"type": "string",
"minLength": 1,
"maxLength": 256,
"description": "Optional transcription model (default gpt-4o-transcribe; whisper-1 and gpt-4o-mini-transcribe also work). On Manaurum's voice key only these three are served; any other is 400 model_not_available."
},
"language": {
"type": "string",
"minLength": 2,
"maxLength": 16,
"description": "Optional ISO-639-1 language hint (e.g. 'ru')."
},
"prompt": {
"type": "string",
"maxLength": 4000,
"description": "Optional vocabulary-biasing prompt (names, domain terms) passed through to the transcription API."
}
}
}
{
"type": "object",
"required": [
"key",
"content_type",
"size_hint"
],
"additionalProperties": false,
"properties": {
"key": {
"type": "string",
"minLength": 1,
"maxLength": 1024
},
"content_type": {
"type": "string",
"minLength": 1,
"maxLength": 255
},
"size_hint": {
"type": "integer",
"minimum": 0,
"maximum": 52428800,
"description": "Exact byte length of the body you will PUT. Signed into the URL — a body of any other size is rejected by the object store."
},
"expires_in": {
"type": "integer",
"minimum": 60,
"maximum": 3600
}
}
}
os.http.fetchv1
app
Input schema
json
{
"$schema": "https://json-schema.org/draft/2020-12/schema",
"type": "object",
"properties": {
"url": {
"type": "string",
"format": "uri",
"minLength": 1
},
"method": {
"type": "string",
"enum": [
"GET",
"POST",
"PUT",
"DELETE"
]
},
"headers": {
"type": "object"
},
"body": {
"type": "string"
},
"body_base64": {
"type": "string",
"maxLength": 7000000,
"description": "Base64-encoded binary request body. Mutually exclusive with `body`. Max ~5 MB decoded, matching the response cap."
},
"response_format": {
"type": "string",
"enum": [
"text",
"base64"
],
"description": "How to return the response body: `text` (default, UTF-8 with replacement — historical behaviour) or `base64` (lossless, for binary downloads; the body is returned in `body_base64` and `body` is empty)."
},
"timeout_ms": {
"type": "integer",
"minimum": 1,
"maximum": 30000
}
},
"required": [
"url"
],
"additionalProperties": false,
"not": {
"required": [
"body",
"body_base64"
]
}
}