Click to Call (C2C) Service APIs

• Public endpoint for email/password authentication.
• Returns bearer access and refresh tokens on success.

• Requires bearer access token.
• For provider Click2Call, did is required. The backend sends the call to the provider in the background.
• For voice-agent calls, pass a voice agent ID such as AGT_E882B100 as agent_number and put call infields in agent_config. did is not needed. See Voice-Agent Calls (AGT_) below.
• If callback_url is provided, the backend posts call and transcript lifecycle events to that URL as the call progresses.
• If callback_url is not provided, use the recent logs and transcript endpoints to poll status.

• Requires bearer access token.
• Use this endpoint to monitor many calls without polling every call_id individually.
• When transcript_status is ready, each call can include transcript data. Otherwise transcript is null.
• recording and recording_url are signed recording URLs when available. Before provider callback or upload, they can be pending.

• Requires bearer access token.
• Polling may return transcript: null with transcript_status: pending while processing.
• If processing fails, the API still returns 200 with transcript_status: failed and the failure reason in data.error.
• Phone numbers are masked in this endpoint response.

• refresh_token is sent as a query parameter, not in the request body.
• The refresh token itself is not rotated. The same refresh token is returned and stays valid until it expires.
• The refresh token must carry the session it was issued with. Tokens issued outside a login session are rejected.

• Requires bearer access token.
• Only numbers assigned to you whose metadata type is c2c are returned. Voice-agent numbers are not listed here.
• Use this endpoint to discover the did values you can pass to POST /api/calls/click2call.
• active_channels is counted from live call records, so it reflects real concurrency rather than a stored counter.
• Stale channel reservations are swept before the response is built, so the counts are current at read time.
• Up to 500 numbers are returned. There is no pagination on this endpoint.

• Requires bearer access token.
• pool_total_channels is your total concurrent capacity. It comes from your C2C batch totals, falling back to the sum of purchased channels across assigned C2C numbers.
• available_channels is allocated_channels minus active_channels for that DID, floored at zero.
• pool_active_channels is always your account-wide live channel count.
• If you have never saved an explicit allocation map, the backend seeds one from purchased channel values. Save an explicit map with PUT to normalize the configuration.

• Requires bearer access token.
• This is a full replacement. Omitted DIDs drop to zero allocated channels, so always send your complete map. Use PATCH to change only some DIDs.
• Identify each DID by either did or phone_number_id. Sending both on one item is rejected; sending neither is rejected.
• No DID may appear twice in one request.
• Each allocated_channels must be between 0 and that DID max_channels, and may not drop below the DID current active_channels.
• The sum of all allocated_channels must not exceed pool_total_channels.
• The response body is the same shape as GET /api/mobile-numbers/channel-allocation.

• Requires bearer access token.
• Use this instead of PUT when you only need to retune a few DIDs and do not want to resend the whole map.
• The same identification, max_channels, active-channel, and shared-pool rules as the PUT endpoint apply.
• The response body is the same shape as GET /api/mobile-numbers/channel-allocation.

• Requires bearer access token.
• Poll this before a burst of calls to confirm you have free channels, rather than relying on the warning in the click2call response.
• Pass did to narrow the numbers array to one DID. The pool counters stay account-wide either way.
• The did field in each numbers item is currently always null. Use phone_number_id to identify the DID, or read the number from GET /api/mobile-numbers/channel-allocation.

• Requires bearer access token.
• Only provider Click2Call transcripts can be retried. Voice-agent (AGT_) calls are rejected with 409.
• The call must have been created with transcript: true, and a stored recording must be available.
• Retrying resets transcript_status to pending and clears any previously sent transcript.ready, transcript.failed, and transcript.disabled callback events, so those events can be delivered again.
• Processing runs in the background. Poll GET /api/calls/{call_id}/transcript or wait for the callback.

• Requires bearer access token.
• Identify the DID with either phone_number_id or phone_number. Sending both is rejected; sending neither is rejected.
• Send callback_url: null to clear the webhook for that DID.
• transcript is a per-DID setting: it decides whether inbound recordings on that number are transcribed. It is sent on every request and defaults to false when omitted.
• When an inbound callback URL is configured, the terminal call event is held until the recording is stored in S3, so the callback carries a usable recording link.
• The DID must already be assigned to a user before a callback URL can be set.

• Requires bearer access token.
• Use configured_only=true to see just the DIDs that already have a webhook, which is the quickest way to audit inbound coverage.
• callback_url is null for DIDs with no webhook configured.
• Up to 500 numbers are returned. There is no pagination on this endpoint.

• Most integrations never call this endpoint. It is the provider-facing ingest hook; consume the outcome through your callback_url or the recent-logs API instead.
• It is documented here because it is the step that turns a queued call into a terminal status, and because partners running their own telephony can post results into it.
• Unlike the other endpoints on this page, this route is served at the platform API root and is not behind the developer route prefix.
• Accepts application/json, application/x-www-form-urlencoded, and multipart/form-data.
• Field names are case and spacing tolerant. "Customer Number", "customer_number", "Agent Number", "agen_number", "DID", "Virtual Number", and "Hangup Cause" are all accepted.
• customer_crm_id carries the Indus call_id and is how the provider result is matched back to the original Click2Call request.
• Ingest is idempotent. A repeated post for a call already processed updates the stored record but does not re-run the recording pipeline or resend a delivered callback event.
• A terminal status releases the DID channel reservation held by that call.

• Voice-agent mode is used when agent_number starts with AGT_ (uppercase). Any other value uses the provider flow.
• did is not required and is ignored. The call is placed from the phone number assigned to the agent for outbound calls, or the platform default when none is assigned.
• The agent must belong to your account, or you must be a member of the agent's team.
• The request waits up to about 15 seconds for the customer to answer. data.status is success if the customer answered, or failed if not. It is never queued.
• Calls use the agent owner's credits and the purchased channels of the agent's outbound number. Click2Call DID channel allocation does not apply.
• The stored call has call_type: agent_telephony. Its status changes from success to completed when the call ends. Filter these calls with GET /api/calls/recent?call_type=agent_telephony.
• transcript_language is ignored. With transcript: true, the transcript is the voice agent's own conversation log.

1. Define each infield on the agent: in the dashboard (Agent tab, Call Infields), or with POST /api/agents/{agent_id}/call_infields (see Call Infields). The field_name you define, e.g. customer_name, is the key you send on each call. Defining fields is recommended so every integration uses the same names; substitution itself works for any key you send.
2. Use the field in the agent's system prompt or first message as a placeholder with single curly braces, e.g. {customer_name}. Make sure that prompt is in the agent's current (published) config.
3. On each click2call request, send the values as flat keys in agent_config. Key names must match the placeholders exactly (case-sensitive).
4. When the call connects, the agent replaces each placeholder with the value you sent for that call.

• Do not nest the values under call_infields. Unlike POST /api/livekit, this endpoint does not unwrap a nested object.
• If a placeholder has no matching key, the agent says it as-is (it speaks the text {customer_name}).
• Values are inserted as text, so format dates and amounts the way the agent should say them.
• Do not put secrets in agent_config.

The platform sets these keys itself. Do not use them as infield names: agent_id, call_id, phone_number, from_number, phone_number_id, sip_call_to, room_id, voice_room_id, metadata, scheduled_time, timezone.

• If the call does not connect: call.failed with error Voice-agent call could not be connected.
• With transcript: false: call.completed, then transcript.disabled, after the call ends.
• With transcript: true: call.completed, then transcript.ready once the transcript and recording are linked. If they are not available within 30 minutes of the call ending, transcript.failed is sent instead.
• The end of the call is checked about every 10 seconds, so events can arrive up to about 10 seconds after hang-up.

• Exceeding a DID allocation or the shared pool does not reject the call. The request is accepted and data.warning explains which limit was crossed.
• Possible warnings: Allocated channel limit exceeded for this DID; call accepted, No channels allocated for this DID; call accepted, and Client channel pool exhausted; call accepted. Multiple warnings are joined with |.
• Treat a non-null warning as a signal to slow down or rebalance allocations, not as a failure.
• Check GET /api/mobile-numbers/channel-allocation/utilization before a burst if you need to stay strictly inside your limits.
• A channel reserved by a call the provider never reports on is freed automatically after 5 minutes, and the call moves to temporaryfailed.

• Answered and NoAnswered are provider-style status values and should be treated as final call results.
• queued is the initial state before a final provider callback is received.
• call.completed is a callback event name, not the same thing as a stored call status value.
• A call that reaches any terminal status releases the DID channel it reserved. Terminal statuses are Answered, NoAnswered, afthrs, failed, temporaryfailed, and completed.
• temporaryfailed is provisional. If the provider reports late, the record is reprocessed and the corrected outcome is delivered to your callback_url.
• Answered calls also carry hangup_cause from the provider, for example ANSWER, BUSY, CANCEL, or NOANSWER.

1. Your server calls POST /api/calls/click2call with callback_url.
2. The API returns call_id immediately. Store it.
3. The backend sends the outbound call request to the provider.
4. The provider later calls the backend with call status and recording details.
5. The backend stores the recording and starts transcript processing when transcript is true.
6. The backend POSTs lifecycle events to your callback_url.

• Your callback_url must be publicly reachable from the backend. Do not use localhost for production calls.
• The callback is sent in the background. Your create request does not wait for the call, recording, or transcript.
• The callback data object uses the same compact shape as GET /api/calls/{call_id}/transcript.
• Return a 2xx response from your webhook to mark the event as delivered.
• If no callback_url is supplied, no webhook is sent. Use GET /api/calls/recent or transcript polling.
• Each event is delivered once per call. Retrying a transcript with POST /api/calls/{call_id}/transcript/retry re-arms the transcript events so they can be delivered again.
• Inbound calls use the per-DID webhook set with POST /api/calls/inbound-callback-url instead of a per-call callback_url.