Skip to main content

​
Knowledge bases

All endpoints live in app/api/knowledge-bases. They require a verified user (auth middleware) and in most cases requirePhoneVerification: true.

​
GET /api/knowledge-bases

Returns every knowledge base (id, name, description, timestamps).

​
PUT /api/knowledge-bases

Body { name: string, description?: string }. Creates a new KB shell with empty metadata and returns the inserted record.

​
GET /api/knowledge-bases/{id}

Loads the KB plus a list of uploaded documents (kb(...).listDocuments()).

​
PATCH /api/knowledge-bases/{id}

Body { name?: string, description?: string }. Updates metadata only.

​
DELETE /api/knowledge-bases/{id}

Removes the KB and iterates through all remote documents to delete them from the vector store (kb(...).deleteDocument). Use with caution—deletion is immediate.

​
Upload files

POST /api/knowledge-bases/{id} accepts multipart/form-data: 
------form
Content-Disposition: form-data; name="files"; filename="faq.pdf"
Content-Type: application/pdf
Large text-based files are chunked into ~200 KB slices to keep embeddings efficient. The response is the list of created documents. POST /api/knowledge-bases/{id}/links body { urls: string[] }. The server fetches each URL (4 concurrent downloads, 15 s timeout), converts the response into File objects, batches them (5 at a time), and calls kb(...).createDocuments. Failed downloads are skipped but logged. POST /api/knowledge-bases/{id}/search body { text: string }. Returns up to 5 matches from the KB’s vector store with their similarity scores and document metadata. Handy for grounding prompts or tool calls.

​
Delete a specific document

DELETE /api/knowledge-bases/{id}/{documentId} removes one document by numeric ID.

​
Personas

Routes under app/api/persona expose persona definitions stored in lib/get-persona.ts.

​
GET /api/persona/{id}

Returns the persona JSON (voice instructions, description, provider-specific settings). Useful when composing call payloads or persona previews.

​
GET /api/persona/list-external

If an ElevenLabs API key is configured (ELEVENLABS_API_KEY), this returns all available ElevenLabs voices converted into Telentir persona objects. When the key is missing or forbidden, the list is empty.

​
POST /api/persona/{id}/play

Generates an encrypted audio preview for a persona so you can render “listen” buttons without exposing raw audio. Body:
FieldDescription
serverWhich of your servers should decrypt the response.
jwtSigned payload verified against the same server. It must decrypt to { use: "persona-preview", textBase64?: string }.
If textBase64 is omitted, Telentir previews the persona’s default description. Depending on persona.settings.type, Telentir either calls OpenAI’s gpt-4o-mini-tts or ElevenLabs’ TTS API and then encrypts the audio buffer with your server keys. The JSON response contains { header, content } ready to decrypt via Server.decrypt.

​
Integrations (Calendly & friends)

app/api/integrations stores third-party credentials (currently Calendly) used by the orchestrator and integration functions surfaced in live calls.

​
GET /api/integrations

Lists all integrations for the workspace. Token-like fields are masked on output (ABCD...1234) to avoid leaking secrets.

​
PUT /api/integrations

Body { type: string, data: Record<string,string> }. For Calendly, supply:
FieldMeaning
accessToken / apiKey / token / patAny of these keys is accepted; the resolver normalizes to accessToken.
eventTypeUrl / eventType / eventTypeSlugIdentifies the scheduling template to use.
The backend validates the credentials (resolveCalendlyIntegration) and stores a sanitized payload containing eventTypeUri, eventTypeSlug, eventTypeName, userUri, and optional schedulingUrl.

​
DELETE /api/integrations/{id}

Revokes a single integration record. Once saved, integrations are exposed through lib/integration-functions.ts and can be attached to call sessions (see /api/session/call).