> ## Documentation Index > Fetch the complete documentation index at: https://docs.thinnest.ai/llms.txt > Use this file to discover all available pages before exploring further. # Delete Call Content > Deletes ONE call's recording and transcript, and keeps its billing record. Gone: the recording (ours, the phone provider's copy of a call on a number we rent, and any copy made for model training), this call's own turns in its conversation (other calls and chats with the same person stay), its summary, its collected fields, and the summary handed on with a warm hand-off. Kept: the call itself — when it happened, who it was with, how long it lasted, what it cost and `billing`. Afterwards Get Call answers with `recording: null`, an empty `transcript`, `contentDeleted: true` and `contentDeletedAt`. A recording that cannot be deleted at that moment is queued and retried every night until it is gone, and counted in `recordingsPending`. For a call you placed that was retried, every attempt's content goes. This cannot be undone. This is not Cancel Call: `DELETE /calls/{id}` still calls a call off. Safe to repeat — again answers 200 with the first deletion's time. Needs a **full** key; a developer reaches a customer's call with `Thinnest-Workspace`. ## OpenAPI ````yaml /api-reference/openapi.json delete /calls/{id}/content openapi: 3.1.0 info: title: ThinnestAI API version: 1.0.0 description: >- Build voice and WhatsApp agents, place calls, message customers and run a platform of your own customers on one API. servers: - url: https://app.thinnest.ai/api/v1 description: Production security: - bearerAuth: [] tags: - name: Agents description: >- An agent is one assistant with its instructions, model, languages, website widget and call settings — the same thing the console's Agent page builds. Create one here, then give it knowledge, actions and channels. - name: Knowledge description: >- The documents an agent searches and cites when it answers — a web page or pasted text, split into passages and indexed. Use it for facts (prices, policies, opening hours) rather than putting them in the instructions. - name: Actions description: >- Calls to your own HTTPS API that an agent can make mid-conversation, with arguments it fills from what the customer said — how an agent uses data that stays in your own system. Create one, test it, then switch it on. - name: Tools description: >- Everything else an agent can do besides answer: its built-in tools, the MCP tool servers it is connected to, and the accounts it works in (calendar, spreadsheets, CRMs, records kept here, helpdesks). Every tool starts off. - name: Channels description: >- Where an agent answers: the website widget, WhatsApp, Telegram and calls. Read an agent's channels here and connect a Telegram bot; WhatsApp is connected in the console, because Meta's sign-up runs in the business owner's own browser. - name: Calls description: >- A call is one phone, web or WhatsApp conversation with an agent — one you placed through the API, a campaign's, or one a customer made to you. Place calls here, one at a time or in a batch, and read each one back with its summary, collected details and transcript. - name: Post-call description: >- What happens after a call ends: the outcome, a short summary and the details the agent collects. Read a call's analysis here, and set what an agent does after every call. - name: Recordings description: >- A call's audio, for agents that record their calls. Fetch it with your key, or hand a CRM user the signed link a call report carries — that one needs no key. - name: Conversations description: >- A conversation is one customer's thread with your workspace on one channel — WhatsApp, the web widget, Telegram, voice or email. Read threads and their messages (including to catch up on a webhook you missed), answer a WhatsApp customer from your own agent platform, and resolve, take over or assign threads as the inbox does. - name: Messages description: >- Send one approved WhatsApp template to one customer, because something happened in your own system — an order confirmation, a delivery update, an appointment reminder. No agent is needed: this is how your software starts a WhatsApp conversation. - name: One-time codes description: >- Deliver a login or verification code you generated over WhatsApp, without handling templates yourself, and set up the authentication templates it sends with. WhatsApp writes these messages — "*394812* is your verification code." — so you choose only the expiry, the security line and the button. - name: Contacts description: >- The people your workspace knows — kept in step with your CRM by phone number and your own `externalId`, with marketing consent recorded alongside where it came from. - name: Do not call description: >- Numbers your agents must never ring. Every outbound call refuses a number on this list — Create Call answers `409`, a batch refuses that entry, a calling campaign skips the person and records why, and a booked callback is given up — while inbound calls from it are answered as usual. The same list as **Contacts → Do not contact** in the console, and the one your voice agent adds to on its own when a caller asks not to be called again (in any language), which also sends the `contact.opted_out` webhook event. - name: Events description: >- Tell us something happened in your system — `order.delivered`, `trial.ended` — and the sequences and tags you set up in the console decide what follows. Your code says what happened; the console says what happens next. - name: Templates description: >- WhatsApp message templates: the pre-approved messages your business may send to a customer first, or outside the 24 hours after they last wrote. A template is saved as a draft, sent to WhatsApp for review, and can be sent (Send Message, Create Campaign) only once it is `approved`. Templates are named, not numbered, and one name can exist in several languages. - name: Campaigns description: >- A campaign sends one thing to many people. It is either a WhatsApp **broadcast** (`kind: "whatsapp"`), which sends one approved template to everyone in its audience, or a **calling campaign** (`kind: "voice"`), where an agent phones each person from its own number within the hours you set. Who is in the audience is decided by consent, exactly as in the console, and every campaign is created as a draft that you then launch, schedule, pause, resume or cancel. - name: Sequences description: >- A sequence sends one person several approved WhatsApp templates over days — "your order shipped", then "how was it?" two days later — and stops by itself when they reply, are tagged, or an event says so. People join by tag, by event, or one at a time through Enrol in Sequence. Sequence ids are bare uuids. - name: Forms description: >- A WhatsApp form is a screen your customer fills in without leaving the chat — a delivery address, an appointment request, a short survey. Build it here, publish it to your own WhatsApp Business Account, then send it from a template's button, an agent, the inbox or Reply to Conversation. - name: Phone numbers description: >- The numbers your agents answer and call from. A workspace holds them two ways: numbers it **rents** here, charged every month from your balance, and numbers it **brings** from its own carrier account, which we never bill. Each number is answered by at most one agent (its line) and may be lent to one agent to call out on. A number is named by the number itself, in any spelling: `918041234567`, `+918041234567` and `%2B918041234567` in a path are the same number. - name: SMS description: >- Your own SMS provider and your own DLT registration, so your agents can text customers in India — during a call (the `send_sms` built-in tool) and after it (an agent's post-call `followUp`). Every business SMS in India must match a template approved on your DLT portal and go out under your six-letter sender header; your provider bills you for each one, and nothing is charged to your wallet here. Providers: `msg91`, `exotel`, `gupshup` (Enterprise SMS), `infobip`, `vonage`, `fast2sms`. On your DLT portal, add your provider as your telemarketer (the PE–TM chain) or every send is rejected. - name: Voices description: >- The voices an agent can speak with, as the console's voice picker lists them. An item's `id` is what an agent's `voice.voice` takes. - name: Voice clones description: >- Your workspace's own cloned voices, made from a short recording with the speaker's consent. A clone is listed with the Studio voices and can be put on an agent like any other voice. - name: Models description: >- The language models an agent can answer with, as the console names them. An item's `id` is what an agent's `model` takes. - name: Bring your own keys description: >- Run your agents on your own provider accounts and pay a flat platform fee instead of our usage rates. Bring all three keys (speech-to-text, LLM, voice): ₹0.02 a chat reply and ₹1 a call minute, all-in. Or bring only your voice key (scope voice): our speech-to-text, model and telephony, a call billed at the voice-only BYOK rate (₹1.50 / $0.02 a minute by default), which includes our model — so calls run only on Prana, Prana [Voice], GPT-OSS 120B or another model GET /api/v1/models marks `voiceOnlyByok` (an agent on any other model runs on Prana on calls) — chat replies as usual, on any model. Or bring your language-model key and your voice key (scope llm_voice): we keep the speech recogniser, a call is a flat ₹1 a minute like all three keys, with no model limit, and a chat reply is the same flat fee. Or bring only your language-model key (scope llm): we keep the speech recogniser and the voice, and a call is billed exactly like any other call on the voice plan the agent speaks with — Standard, Premium or Studio, with your plan's discount — whatever model your key runs, with no model limit; chat replies are billed at the same flat own-keys fee as under all three keys. There is no fallback to our providers when one of your keys fails. The switch and the default `scope` are per workspace, and each agent follows them by default. Each agent can choose for itself (Create Agent, Update Agent `byok`): `null` follows the workspace, `"off"` runs that one agent on our voices and models at our normal pricing — chat replies billed normally, the voice-only model limit not applied — and a scope (`all`, `voice`, `llm` or `llm_voice`) runs it on that scope's keys, so one workspace with an LLM key and a voice key can have one agent on `llm` and another on `llm_voice`. A scope is refused (409, naming the missing key) unless you hold verified keys for it, and a key an agent's scope needs cannot be removed while BYOK is on. If a key is gone when a call starts, the agent falls back to the workspace's scope, never fails the call. A call is priced and run on one stack: the agent's setting is read once when the call or reply starts. - name: Webhooks description: >- Endpoints an agent's events are sent to as they happen — a CRM, a sheet, Slack, Zapier, or a helpdesk email address. Each delivery is a signed JSON `POST` of `{ "id", "event", "sentAt", "data" }`. **Headers on every attempt** (not sent to an email destination, which gets one ticket email per event) | Header | What it is | |---|---| | `x-thinnest-signature` | `sha256=` under the endpoint's signing secret. | | `x-thinnest-signature-v2` | `sha256=.">` — the delivery time, signed. | | `x-thinnest-delivered-at` | When this attempt left (ISO 8601). | | `x-thinnest-event-id` | `evt_…`, the same as the body's `id` — stable across every retry and re-send. | | `x-thinnest-attempt` | `1` for the first attempt, then `2`, `3`, … | **Retries.** A delivery that fails — any answer other than 2xx, a timeout after 10 seconds, or an address we cannot reach — is tried again after 1 minute, 5 minutes, 30 minutes, 2 hours and 6 hours (six attempts over about 8½ hours). Every attempt sends the **same bytes**, so `sentAt` is when the event happened, not when the attempt left. Payloads are kept for 7 days; within that time any failed delivery can be sent again with Redeliver Webhook Events. **Dedupe** on `x-thinnest-event-id` (or the body's `id`): a retry after a timeout can reach you after you already processed the first attempt. Answer 2xx quickly and do the work afterwards. **Freshness.** To refuse replayed requests, verify `x-thinnest-signature-v2` and reject an `x-thinnest-delivered-at` more than five minutes from your clock. Do not check freshness on `sentAt`: a retry's `sentAt` is hours old by design. **Switch-off.** An endpoint is switched off after five events in a row that each failed every attempt — single failed attempts that later succeed do not count. Events that happen while it is off are recorded as not sent; switch it back on and re-send them. **Ids in `data`** are the prefixed ids the API returns and takes: `conversationId` is `conv_…` (the conversation endpoints take it), `campaignId` is `cmp_…`, and a call is `out_…`, `sch_…` or its call reference (Get Call takes it). The `WebhookPayload` schema has a full example of every event's body. The events an endpoint can subscribe to: | Event | When it fires | |---|---| | `lead.captured` | The agent collected a name, email or phone from somebody who showed real interest. | | `conversation.escalated` | The agent handed a conversation to a person — asked for, or because it could not answer. | | `conversation.resolved` | A teammate marked a conversation done. | | `call.completed` | A call the agent placed or answered ended — how it ended and how long it lasted, sent the moment it ends. | | `call.analysed` | A call's results are ready: its summary, the details it collected, the transcript, a recording link and what the call cost (`costMicro`, `currency`) — the same shape as Get Call. | | `campaign.finished` | Every person on a broadcast or a calling campaign has been reached, or given up on. | | `contact.opted_out` | During a call, somebody asked not to be called again; their number (`data.phone`) is now on the do-not-call list; `data.source` is `"call"`; `data.reason` is the words they used, as recorded, or null. | | `call.problem` | During a call: the agent could not answer, your own model key failed, a reply was very slow, or the voice connection failed. Sent once per kind per call. | | `contact.erased` | A contact you deleted is gone completely, recordings included — at once, or after the nightly retry when a recording could not be deleted straight away (`recordingsPending`). Sent once. | | `byok.changed` | Your own keys changed: a key was added, replaced, removed or failed its check, or your own keys were switched on or off or set to cover different jobs. | **`call.problem`** is sent while the call is still going on, at most once per kind per call: `data` is `{ "callId", "agentId", "kind", "detail", "at" }`, where `callId` is the id Get Call takes and `kind` is `no_answer_from_model` (the agent could not get an answer and apologised to the caller), `byok_failed` (your own model key failed), `slow_reply` (a reply took more than 5 seconds to start) or `voice_failed` (the call's voice connection failed). `detail` is a sentence to show a person. **Calls handed to another agent.** `lead.captured` and `conversation.escalated` carry `data.agentId` (the agent that ran the tool) and `data.callId` (the id Get Call takes; null off a call). On a call handed to another agent the call-level events — `call.completed`, `call.analysed`, `call.problem` — stay with the agent the call started with, and `handedTo` on `call.analysed` and Get Call lists the others, so `callId` joins them. Each `handedTo` entry says whether the hand-off was `warm` or `cold` (`mode`) and, when warm, the `summary` the next agent was given. A hand-off to another agent is not a hand-over to a person and sends no `conversation.escalated`. **Usage and billing per call.** `call.analysed` carries `data.usage` and `data.billing`, the same objects as Get Call's: `inputTokens`, `outputTokens` and `cachedInputTokens` the call's replies used (your own model key's too), `inputTokensEstimated` (true when part of the input count was estimated because the provider reported none) and `ttsCharacters` (text sent to the voice, including text sent before the caller interrupted); and `billing` — whose keys the call ran on (`keys`), the voice plan it was priced on (`voicePlan`), its per-minute rate (`ratePerMinuteMicro`) and the seconds it was charged for (`billedSeconds`). **Workspace events.** `contact.erased` and `byok.changed` are about the whole workspace, not one agent. They go to every endpoint in the workspace that subscribed to them, whichever agent it was added on — once per address, so three agents posting to the same URL still receive one delivery — and to a developer's every-customer endpoints (`includeCustomers`) for an event in one of its customers. An endpoint subscribed to every event (`["*"]`) receives them too. Neither carries personal data or key material: - `contact.erased`: `data` is `{ "contactId", "erasedAt" }` — the `cust_…` id Delete Contact was given and when it was erased. Sent exactly once: straight after Delete Contact when it answered 204, or from the nightly retry once the last recording it reported as `recordingsPending` is gone (a recording still being written can take up to three nights to be confirmed). - `byok.changed`: `data` is `{ "change", "kind", "provider", "byokEnabled", "scope" }`. `change` is `key_added`, `key_replaced`, `key_removed`, `key_failed_verification` (the provider refused a key you tried to save; nothing was saved), `byok_enabled`, `byok_disabled` or `scope_changed`. `kind` (`stt`, `llm` or `tts`) and `provider` name the key for the four key changes and are null for the others. `byokEnabled` and `scope` are where the workspace stands after the change. A change of a key's model is not sent, and saved keys are not re-checked on a schedule: a model key that stops working on a call shows up as `call.problem` (`byok_failed`). **Call quality.** `call.completed` and `call.analysed` carry `data.quality`, the same object as Get Call's `quality`: reply times (`responseMs` from the caller falling silent to the agent's first audio, `firstTokenMs` our side of it — each median, p90 and max in milliseconds), long silences, interruptions, and the turns the agent apologised for (`modelErrors`) or a backup model answered (`rescues`). It is null for a call that was not measured, and on `call.completed` when the call ended before its figures arrived — `call.analysed` and Get Call have them by then. - name: Usage description: >- Read what your workspace used and spent — replies, calls, WhatsApp, the balance history and invoices — the same figures as the console's Usage and Billing pages. All reads. - name: Analytics description: >- The Dashboard's counts and each agent's Analytics page, for building a dashboard of your own that agrees with the console. - name: Customers description: >- Workspaces you create for your own customers when you build a product on this API. Each customer has its own agents, contacts, knowledge, phone numbers and WhatsApp; your one key reaches any of them by adding `Thinnest-Workspace: ` to a request, and everything they use is charged to you. Manage customers from your own workspace, without that header. - name: Workspace description: >- Your workspace's own settings — its name and time zone — as Settings → General shows them. With `Thinnest-Workspace`, the customer's workspace instead. - name: Projects description: >- Projects group a workspace's agents, and are what a team's access is scoped by. Every workspace has a default project that cannot be deleted; a second project needs a plan with teams and projects. - name: Teams description: >- Teams are groups of members and the projects they can reach. A project granted to no team is open to the whole workspace; granting it to a team restricts it to that team's members (owners and admins always see everything), and removing its last grant opens it up again. - name: Members description: >- The people in your workspace and their roles. An API key acts as an admin, never an owner: it can make people admins or members and remove them, but cannot make anybody an owner, and the last owner can be neither demoted nor removed. - name: Invitations description: >- Invitations to join your workspace, as Settings → Members sends them: an email with a link that works for 14 days. Pending invitations count towards your plan's seats. - name: Notifications description: >- What each kind of notification does in your workspace: shown in the console's feed or muted, and whether it is emailed through the email sender you connected in the console. - name: Saved replies description: >- Sentences your team sends often, offered in the inbox composer as `/shortcut`. - name: Contact groups description: >- Named lists of contacts — "Wholesale buyers", "Pune walk-ins" — the same groups as **Contacts → Groups** in the console. Aim a broadcast or a calling campaign at one with `groups` on Create Campaign. A group narrows who a campaign reaches and never widens it: consent still decides, exactly as for tags. Deleting a group never deletes its contacts. List a group's contacts with List Contacts and `group=grp_…`. paths: /calls/{id}/content: delete: tags: - Post-call summary: Delete Call Content description: >- Deletes ONE call's recording and transcript, and keeps its billing record. Gone: the recording (ours, the phone provider's copy of a call on a number we rent, and any copy made for model training), this call's own turns in its conversation (other calls and chats with the same person stay), its summary, its collected fields, and the summary handed on with a warm hand-off. Kept: the call itself — when it happened, who it was with, how long it lasted, what it cost and `billing`. Afterwards Get Call answers with `recording: null`, an empty `transcript`, `contentDeleted: true` and `contentDeletedAt`. A recording that cannot be deleted at that moment is queued and retried every night until it is gone, and counted in `recordingsPending`. For a call you placed that was retried, every attempt's content goes. This cannot be undone. This is not Cancel Call: `DELETE /calls/{id}` still calls a call off. Safe to repeat — again answers 200 with the first deletion's time. Needs a **full** key; a developer reaches a customer's call with `Thinnest-Workspace`. operationId: deleteCallContent parameters: - $ref: '#/components/parameters/CallIdParam' - $ref: '#/components/parameters/Workspace' responses: '200': description: The content is deleted. `call` is Get Call's answer now. content: application/json: schema: $ref: '#/components/schemas/CallContentDeleted' example: contentDeleted: true contentDeletedAt: '2026-10-15T09:12:40.118Z' recordingsPending: 0 call: null '400': $ref: '#/components/responses/WorkspaceHeaderNotDeveloper' '401': $ref: '#/components/responses/Unauthorized' '403': description: The key is not a full key. content: application/json: schema: $ref: '#/components/schemas/Error' example: error: >- Deleting a call's recording and transcript needs a full API key. '404': description: No call with that id in this workspace. content: application/json: schema: $ref: '#/components/schemas/Error' example: error: No call with that id. '409': description: >- The call is still going on, or it was booked and has not rung (cancel it with Cancel Call instead). content: application/json: schema: $ref: '#/components/schemas/Error' example: error: >- That call is still going on. Delete its content once it has ended. '410': $ref: '#/components/responses/WorkspaceDeleted' '429': $ref: '#/components/responses/TooManyRequests' '503': description: >- Something could not be deleted just now. Nothing is lost by trying again. content: application/json: schema: $ref: '#/components/schemas/Error' example: error: >- Could not delete that call's content just now. Nothing is lost by trying again. components: parameters: CallIdParam: name: id in: path required: true description: >- The call's `id` as Place Call or List Calls gave it: `out_…` or `sch_…` for a call you placed, the call reference for any other. schema: type: string example: out_9f2c1a44-6b0e-4c1d-8a7f-2e5b3c9d1f60 Workspace: name: Thinnest-Workspace in: header required: false description: >- Developers only: the customer workspace this request acts in — its `org_…` id from `POST /customers`. Leave it out to act in your own workspace. Refused with `404` (`Workspace not found.`) when the id is unknown, is not one of your customers, or is not an `org_…` id; `410` when that customer was deleted (restore it with `POST /customers/{id}/restore` until it is erased); `403` when the key belongs to one customer workspace (it cannot act as another); `400` when your workspace is not a developer workspace. `/customers/*` and `GET /customers/usage` are your own workspace's and refuse the header with `400`: call them without it. schema: type: string example: org_3fKq9TzQ1mN8vB2xR7cLpA schemas: CallContentDeleted: type: object required: - contentDeleted - contentDeletedAt - recordingsPending - call properties: contentDeleted: type: boolean const: true contentDeletedAt: type: string format: date-time description: When the content was first deleted — the same on every repeat. recordingsPending: type: integer description: >- Recording files that could not be deleted at that moment and are retried every night until they are gone. call: description: The call as Get Call now answers. oneOf: - $ref: '#/components/schemas/CallReport' - type: 'null' Error: type: object required: - error - code properties: error: type: string description: What went wrong, in a sentence you can show a person. code: type: string enum: - unauthorized - key_scope_read - key_scope_build - workspace_not_found - workspace_deleted - workspace_not_developer - workspace_header_forbidden - rate_limit_resources - rate_limit_family - rate_limit_calls - rate_limit_messages - rate_limit_codes - rate_limit_code_recipient - concurrent_calls - number_daily_limit - rate_limited - opted_out - do_not_call - stop_list_unknown - insufficient_balance - outside_calling_hours - number_never_callable - call_in_progress - call_already_scheduled - calling_not_set_up - carrier_unavailable - agent_not_found - agent_ambiguous - from_number_invalid - agent_on_live_call - phone_number_required - signing_unavailable - invalid_number - number_required - number_claimed_elsewhere - needs_own_carrier_keys - number_move_needs_developer - number_move_same_workspace - number_on_live_call - number_move_needs_business_details - number_move_carrier_account - number_move_unavailable - number_move_in_progress - number_move_legacy_number - number_cannot_forward - forward_destination_not_allowed - forward_loop - template_not_found - template_not_approved - template_language_missing - template_unsendable - code_template_not_found - code_template_ambiguous - whatsapp_not_connected - whatsapp_blocked - whatsapp_not_configured - free_allowance_used - send_failed - plan_required - plan_limit - payment_required - idempotency_in_progress - validation_failed - forbidden - not_found - conflict - gone - unprocessable - internal_error - service_unavailable description: >- A stable machine-readable code. Two refusals that share a status are told apart by it (a `403` for an opted-out customer, for the do-not-call list and for a key that is not a full key). See the table in the API reference under Errors. limit: type: object description: >- On a `429` that hit one of our limits: which one and how big it is. `perMinute` for a per-minute limit; `max` (and `windowSeconds`) for the others. required: - name properties: name: type: string description: >- Which limit: `resources`, `family`, `calls`, `messages`, `codes`, `code_recipient`, `concurrent_calls` or `number_daily`. perMinute: type: integer description: Requests allowed a minute. max: type: integer description: The most allowed, for a limit that is not per minute. windowSeconds: type: integer description: The window `max` is counted over, when it has one. example: error: That agent was not found. code: not_found description: >- Every error body. Branch on `code`, never on the sentence in `error`: the sentence may be reworded, the code does not change. CallReport: description: One call and what it learned — the same object `call.analysed` delivers. allOf: - $ref: '#/components/schemas/CallSummary' - type: object required: - transcript - recording - retryScheduledFor - costMicro - currency - quality - usage - billing properties: transcript: type: array items: $ref: '#/components/schemas/CallTurn' description: This call's spoken turns, oldest first. recording: type: - object - 'null' description: >- A signed link to the audio that needs no key — Get Recording By Signed Link. Null when the call was not recorded. required: - url - expiresAt - ready properties: url: type: string format: uri description: The signed link. expiresAt: type: string format: date-time description: When the link, and the recording, go. ready: type: boolean description: False while the audio has not landed yet. retryScheduledFor: type: - string - 'null' format: date-time description: >- When the next try will ring, if this one did not reach them and another is queued. costMicro: type: - integer - 'null' description: >- What this call cost the workspace, in millionths of `currency` (6370000 is 6.37) — the figure `GET /usage/calls` gives, at a white-label reseller's price for its clients. Null until the call is settled. Settlement runs before `call.analysed` is sent, so the webhook normally carries it, but it is null there too when the call has not settled; read Get Call later for the figure. currency: type: - string - 'null' enum: - INR - USD - null description: >- The currency `costMicro` is in — the workspace's. Null only for a booked call that has not rung. quality: $ref: '#/components/schemas/CallQuality' usage: $ref: '#/components/schemas/CallUsage' billing: $ref: '#/components/schemas/CallBilling' CallSummary: type: object description: A call, without its transcript and recording. required: - id - callRef - reference - metadata - attempt - status - direction - phone - from - agent - scheduledFor - startedAt - answeredAt - endedAt - seconds - hangup - variables - summary - fields - analysedAt - handedTo - forwardedTo - contentDeleted - contentDeletedAt - error properties: id: type: string description: >- The id to open it by: `out_…` / `sch_…` for a call you placed, the call reference otherwise. callRef: type: - string - 'null' description: >- The reference of the attempt that actually rang. Null on a call that has not rung. reference: type: - string - 'null' description: Your `reference` from Place Call. metadata: type: - object - 'null' additionalProperties: type: - string - number - boolean - 'null' description: Your `metadata` from Place Call, untouched. attempt: type: integer minimum: 1 description: 1 for the first try, 2 for the first retry. status: type: string enum: - scheduled - ringing - connected - completed - missed - failed - cancelled description: >- Where the call is. `scheduled` and `cancelled` only appear on calls you placed for later. direction: type: string enum: - inbound - outbound description: Who rang whom. phone: type: - string - 'null' description: The customer's number. from: type: - string - 'null' description: >- Your number the call rang from. Null until it rings, unless the request chose one, and always null on a call that came in. agent: $ref: '#/components/schemas/CallAgentRef' scheduledFor: type: - string - 'null' format: date-time description: When a booked call will ring. Null once it has. startedAt: type: - string - 'null' format: date-time description: When it started ringing. answeredAt: type: - string - 'null' format: date-time description: When it was picked up. endedAt: type: - string - 'null' format: date-time description: When it ended. seconds: type: integer minimum: 0 description: >- How long it lasted. On a call forwarded to a phone (`forwardedTo`) this is how long the caller was connected - answer to hang-up - even when nothing was billed (a number you brought bills nothing for a forward). 0 for a forward nobody picked up (`status` `missed`). hangup: type: - string - 'null' enum: - answered - no_answer - busy - rejected - unreachable - cancelled - voicemail - before_agent - failed - not_placed - null description: >- Why it ended. Null while live; `not_placed` for a booked call that never rang. variables: type: object additionalProperties: type: string description: The `{{name}}` values the call used. summary: type: - string - 'null' description: Two or three sentences about the call, when summarised. fields: type: object additionalProperties: true description: >- The collected details, keyed by field name. Empty when nothing was said. analysedAt: type: - string - 'null' format: date-time description: >- When the summary and fields were written. Null means they may still arrive. handedTo: type: array description: >- Every agent the call was handed to mid-call, oldest first (set on the agent's Actions page, under Hand calls to another agent). Empty when it stayed with `agent`, which is always the agent the call was placed or answered as. A hand-off to another agent is not a hand-over to a person: it sends no `conversation.escalated`. items: type: object required: - agent - from - at - mode - summary properties: agent: $ref: '#/components/schemas/CallAgentRef' from: $ref: '#/components/schemas/CallAgentRef' at: type: string format: date-time description: When it was handed over. mode: type: string enum: - warm - cold description: >- How it was handed over. `warm`: the agent got a summary and the conversation. `cold`: it started without the conversation and greeted the caller afresh. Calls handed over before warm and cold existed read `warm`. summary: type: - string - 'null' description: >- On a warm hand-off, the summary the next agent was given: written by the previous agent, or built from the caller's last words when it wrote none. Up to 300 characters. Null on a cold hand-off. forwardedTo: type: - string - 'null' description: >- The phone this call was forwarded to, when the number forwards every call (`forwardTo` on the phone number) and no agent answered it. Null on every other call. A forwarded call has no transcript, summary or fields. contentDeleted: type: boolean description: >- True once the call's recording and transcript were deleted — by Delete Call Content, or by the workspace's `callRetentionDays`. The summary, fields, transcript and recording are then empty; when it happened, how long it lasted and what it cost stay. contentDeletedAt: type: - string - 'null' format: date-time description: When its content was deleted; null while it is kept. error: type: - string - 'null' description: Why a booked call could not be placed, in words. Null otherwise. CallTurn: type: object required: - speaker - text - at properties: speaker: type: string enum: - customer - agent - team description: Who spoke — `team` is a teammate who took the conversation over. text: type: string description: What was said. at: type: string format: date-time description: When the turn was written down. CallQuality: type: - object - 'null' description: >- How the call went, in numbers, written once when it ends. Null for a call that was not measured — a call from before call quality was recorded, a call that never connected, or one whose voice connection never reported. A figure inside is null when the part of the call that measures it did not report. required: - turns - responseMs - firstTokenMs - longSilences - longSilenceSeconds - longestSilenceSeconds - interruptions - modelErrors - rescues properties: turns: type: - integer - 'null' minimum: 0 description: >- Caller turns the agent answered. A turn that got no reply (an echo of the agent's own voice, a turn after it hung up) is not counted. responseMs: type: - object - 'null' description: >- How long the caller waited: from when they stopped speaking to the agent's first audio on the line. Null when it was not measured. required: - median - p90 - max properties: median: type: integer minimum: 0 description: Half the turns were faster than this, in milliseconds. p90: type: integer minimum: 0 description: Nine turns in ten were at least this fast, in milliseconds. max: type: integer minimum: 0 description: The slowest turn, in milliseconds. firstTokenMs: type: - object - 'null' description: >- Our side of that wait: from the turn reaching the model to its first word. Null when it was not measured. required: - median - p90 - max properties: median: type: integer minimum: 0 description: Half the turns were faster than this, in milliseconds. p90: type: integer minimum: 0 description: Nine turns in ten were at least this fast, in milliseconds. max: type: integer minimum: 0 description: The slowest turn, in milliseconds. longSilences: type: - integer - 'null' minimum: 0 description: Silences, on either side, at least `longSilenceSeconds` long. longSilenceSeconds: type: - number - 'null' description: >- How long a silence had to be to count: the agent's "are you still there?" time, or 10 seconds when it has none. longestSilenceSeconds: type: - number - 'null' description: The longest stretch where nobody spoke. interruptions: type: - integer - 'null' minimum: 0 description: Times the caller started speaking while the agent was talking. modelErrors: type: - integer - 'null' minimum: 0 description: >- Turns where the agent could not get an answer and apologised to the caller instead. Null when this count was not available for the call. rescues: type: - integer - 'null' minimum: 0 description: >- Turns where the agent's model failed and a backup model answered, so the caller still got a reply. Null when this count was not available for the call. CallUsage: type: object description: >- What the call used: the language-model tokens its replies used, whether the model is ours or your own key's (BYOK), and the characters sent to the voice. Each is null when it was not recorded: a call that never reached the model (a missed call, a forward, a booked call that has not rung). The input count is what the provider reported; where a provider reports none it is estimated from the prompt's length, and `inputTokensEstimated` says so. The call's summary and field extraction after it ends are not counted. required: - inputTokens - outputTokens - cachedInputTokens - inputTokensEstimated - ttsCharacters properties: inputTokens: type: - integer - 'null' description: >- Tokens sent to the model across the call's replies, cached ones included. outputTokens: type: - integer - 'null' description: Tokens the model wrote across the call's replies. cachedInputTokens: type: - integer - 'null' description: >- How many of `inputTokens` the provider served from its cache. Null when the provider reports no cache figure. inputTokensEstimated: type: - boolean - 'null' description: >- True when any reply's input count was estimated from the prompt's length because the provider reported none, so `inputTokens` is partly an estimate; false when every reply's count came from the provider. Null when no tokens were recorded, and for calls before 15 October 2026. ttsCharacters: type: - integer - 'null' description: >- Characters of text sent to the voice to be spoken: every reply, the greeting, fillers and other fixed lines. It includes text sent before the caller interrupted, even if they never heard it, which is what a voice provider charges for, and it is counted on your own voice key (BYOK) too. Null when nothing was sent. CallBilling: type: - object - 'null' description: What the call was billed as. Null for a booked call that has not rung. required: - keys - voicePlan - ratePerMinuteMicro - billedSeconds - currency properties: keys: type: string enum: - 'off' - all - voice - llm - llm_voice description: >- Whose keys the call ran on, as it was set up when the call started: `off` for ours; `all` your speech recognition, model and voice; `voice` your voice only; `llm` your model only; `llm_voice` your model and voice. voicePlan: type: - string - 'null' enum: - standard - premium - studio - null description: >- The voice plan the call was priced on (`off` and `llm` calls). Null for the flat own-key prices (`all`, `voice`, `llm_voice`), for a free trial call, and for calls before 15 October 2026. ratePerMinuteMicro: type: - integer - 'null' description: >- The per-minute price the call was charged at, fixed when it started, in millionths of `currency` (1820000 is 1.82) — at a white-label reseller's price for its clients. A call is charged in 30-second steps (a WhatsApp call by the minute) at half of this each. Null for a free trial call. billedSeconds: type: - integer - 'null' description: >- The seconds the charge was for: the call's length rounded up to its step, or to the phone carrier's block for a call that ended before the agent spoke (charged the carrier's fee only). Null until a charge was taken — a call still live, a free trial call, and calls settled before 15 October 2026. currency: type: - string - 'null' enum: - INR - USD - null description: The currency the rate is in, as `currency` on the call. CallAgentRef: type: - object - 'null' description: >- The agent on the call. Null on a call forwarded to a phone when no agent answered the number (`forwardedTo` is set). required: - id - name properties: id: type: string description: Its `ag_…` id. name: type: - string - 'null' description: Its name. responses: WorkspaceHeaderNotDeveloper: description: >- `Thinnest-Workspace` was sent from a workspace that is not a developer workspace. content: application/json: schema: $ref: '#/components/schemas/Error' example: error: >- Thinnest-Workspace needs a developer workspace. Create a customer first with POST /v1/customers. code: workspace_not_developer Unauthorized: description: >- The key is missing, wrong or revoked. The answer is always the same, whichever it is. content: application/json: schema: $ref: '#/components/schemas/Error' example: error: Unauthorized code: unauthorized WorkspaceDeleted: description: >- The customer named in `Thinnest-Workspace` was deleted. `POST /customers/{id}/restore` brings it back until its erase date. content: application/json: schema: $ref: '#/components/schemas/Error' example: error: >- This workspace was deleted on 2026-10-06 and is erased on 2026-11-05. POST /v1/customers/{id}/restore brings it back until then. code: workspace_deleted TooManyRequests: description: >- Over a per-minute limit. Wait for `Retry-After` seconds. `code` and `limit` say which limit: `rate_limit_resources` (240 requests a minute for the workspace), `rate_limit_family` (2,400 across a developer's customers), `rate_limit_calls`, `rate_limit_messages`, `rate_limit_codes`, `rate_limit_code_recipient`, `concurrent_calls` or `number_daily_limit`. headers: Retry-After: schema: type: integer description: Seconds to wait. content: application/json: schema: $ref: '#/components/schemas/Error' example: error: Over 240 requests a minute. Slow down and retry. code: rate_limit_resources limit: name: resources perMinute: 240 securitySchemes: bearerAuth: type: http scheme: bearer description: >- Your API key (`ta_live_…`) from **Settings → API keys**, sent as `Authorization: Bearer `. Keep it on a server: it can message every customer you have. A key is full, build or read-only; a request its level does not allow is refused with `403`. ```` This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.