CallAgentResource
- analysisType: string
_prompt required - Properties: 1analysisType:
_schema required - createdType: stringFormat: date-time
_at requiredthe date-time notation as defined by RFC 3339, section 5.6, for example, 2017-07-21T17:32:28Z
- idType: stringmin length:1max length:512required
- Properties: 1inputType:
_schema required - instructionsType: stringrequired
The instructions for each call. Use
$nameor${name}to insert a top-level input property. Declare every placeholder ininput_schema.properties. Supply every referenced value in the call'sinputs, even if the schema does not require it. Use$$for a literal dollar sign. Nested property paths are not supported. - isType: boolean
_active required - nameType: stringmin length:1required
- reviewType: enum
_status requiredvaluespending_reviewapproved - submittedType: stringFormat: date-time
_at requiredthe date-time notation as defined by RFC 3339, section 5.6, for example, 2017-07-21T17:32:28Z
- activeType: stringFormat: date-timenullable
_updated _at the date-time notation as defined by RFC 3339, section 5.6, for example, 2017-07-21T17:32:28Z
- additionalType: Additional Participantsnullable
_participants The additional participants calls using this Call Agent may bring on, keyed by each participant's name. Absent, calls using this agent are ordinary two-party calls and a request naming a participant is rejected. Present, a call may supply numbers for these names; the agent still decides during the call when to bring each one on. When it has at least one participant, primary_participant is required. The names agent, assistant, caller, system, and user are reserved, in any letter case. Two names, and primary_participant and a name, must differ by more than letter case, spacing, or character form, such as a full-width letter.
- agentType: booleannullable
_speaks _first Whether calls using this Call Agent should begin speaking as soon as the answered call connects. Null and false preserve the wait-for-caller behavior.
- archivedType: stringFormat: date-timenullable
_at the date-time notation as defined by RFC 3339, section 5.6, for example, 2017-07-21T17:32:28Z
- callType: string
_ended _webhook _url min length:1Format: urinullableHTTPS URL Resia POSTs a call.ended event to each time a call this Call Agent places or answers reaches a terminal status — completed, error, or canceled. The body is a JSON WebhookEvent: {"type": "call.ended", "call": ...}, where call is the same CallDetail that GET /v1/calls/{call_id} returns, built from live data when the event is sent. Resia makes one immediate attempt when the call ends and retries a failed delivery on a fixed schedule (six attempts over about 8.6 hours); any 2xx response is success. The call's id repeats unchanged across retries, so treat a repeated call.id as a duplicate. Treat the complete URL as a credential: delivery authenticity is HTTPS plus URL secrecy, and you may embed your own token in the URL to authenticate Resia's requests. Null clears the webhook and stops pending deliveries.
- descriptionType: stringmin length:1max length:280nullable
- disclaimerType: stringmax length:300nullable
Plain text that Resia speaks word for word, once to each participant when their line connects, before the agent speaks. The agent's voice speaks it, and the participant cannot interrupt it. At most 300 characters. Empty or blank text is stored as null. Null means no disclaimer.
- enableType: booleannullable
_background _noise Whether calls of this Call Agent play a quiet office background sound under the agent's voice. True plays the sound. False plays no sound. Null uses the Resia default, which plays the sound.
- inType: nullable
_call _llm Properties: 2The provider and model used for the live conversation. Choose a provider and model that the gateway accepts. Resia validates the selection when it saves the Call Agent. Null means Resia dynamic (recommended): Resia sends no model, the voice platform selects the model, and no surcharge applies.
- inboundType: string
_input _webhook _url min length:1Format: urinullableHTTPS URL Resia calls to fetch the inputs for a new inbound conversation this agent answers. For a Call Agent, Resia POSTs a call.incoming event — {"type": "call.incoming", "call": {"from_phone_number": , "to_phone_number": }} — when an inbound call arrives. For a Chat Agent, Resia POSTs a chat.incoming event — {"type": "chat.incoming", "chat": {"channel": "sms", "participant_handle": , "agent_handle": }} — when a text starts a new chat; later texts in that chat do not call it. Neither event carries message content. Resia waits up to 30 seconds for a single attempt — no retry. The 200 response body must be the InboundCallInputWebhookResponse envelope, {"inputs": {...}}, at most 1 MiB (1048576 bytes); a bare inputs object is an invalid response. inputs must satisfy this agent's input_schema; it is stored as the call's or chat's inputs and rendered into the agent's instructions. On a timeout, a non-200 status, an unreadable body, or inputs the input schema rejects: a Call Agent whose input schema accepts an empty object answers unpersonalized with empty inputs, and any other Call Agent lets the call ring out; a Chat Agent starts no chat and sends no reply, so a Chat Agent's webhook works as an allow list — answer with inputs for a person you accept (an empty object when the agent needs none) and 404 for anyone else. An agent whose instructions need inputs must have this URL before PATCH /v1/phone-numbers/{phone_number} can assign it to a number. The webhook may be delivered more than once for the same call or person, so the endpoint must be read-only and tolerate duplicates. The URL is not a credential inside Resia: Resia stores it, returns it on reads, and logs it in the clear. Delivery authenticity to your endpoint is HTTPS; embed your own token in the URL if you want to authenticate Resia's requests. Inbound-only: outbound calls and chats always supply inputs in their start request and never consult this URL. Null clears the webhook.
- modelType: enumnullable
_architecture How calls using this Call Agent produce the agent's speech. stt_llm_tts transcribes the other party, writes each reply with a language model, and speaks it with a voice. speech_to_speech runs the whole conversation on one speech-to-speech model, named in speech_to_speech_model. Null means stt_llm_tts, the behavior of every Call Agent before this field existed. A speech_to_speech Call Agent needs a null in_call_llm. Its calls can use a database and custom tools. Its single, orchestration, and inbound calls can bring on additional participants with dial on_agent_request and intro private_first; a batch call dials as a plain call. voices takes only voices of the model's provider: openai for GPT-Live, gemini for Gemini Live, and cartesia or fish for resia-fullplex-1. agent_speaks_first works: with true, the agent speaks first when the call connects. should_leave_voicemail applies to outbound calls: with true, the agent leaves its message after the mailbox beep. A speech_to_speech Call Agent cannot yet answer inbound calls on your own number (SIP trunking).
valuesstt_llm_ttsspeech_to_speech - primaryType: string
_participant min length:1max length:40nullableWhat the primary party — the person at to_phone_number, or the inbound caller — is called in the transcript and in the agent's instructions, e.g. "patient". Required exactly when additional_participants is present, and forbidden otherwise. The names agent, assistant, caller, system, and user are reserved, in any letter case.
- shouldType: booleannullable
_leave _voicemail Whether calls using this Call Agent should leave a message when they reach a voicemail box. Null and false hang up without speaking instead.
- speechType: string
_to _speech _model min length:1max length:64nullableThe speech-to-speech model that runs the calls, for example gpt-live-1. Required when model_architecture is speech_to_speech, and must be null otherwise. Resia validates the name when it saves the Call Agent. GPT-Live, Gemini Live, and resia-fullplex-1 calls speak in one voice from voices, or the model's default voice when voices names none.
- tagsType: TagsProperties: 1
Open key/value labels on this record. Set any keys and any values. Filter a list by one key and value pair. Resia never interprets a tag. Do not put names, phone numbers, or any personal, health, credential, or secret information in a tag: a tag travels in URLs when used as a filter, and URLs appear in ordinary infrastructure request logs.
- timezoneType: stringmin length:1max length:64nullable
IANA time zone name, for example America/New_York. Every call's system prompt states today's date, computed when the call is placed or answered. Without a timezone the date is read in America/New_York and no time is given. With a timezone the date is read in that zone and the prompt also states the local time, for example "6:08 PM". Resia validates the name when it saves the Call Agent. Null clears it.
- toolsType: array of string …64nullable
The names of your organization's tools that calls of this version may use. Resia sends each tool call to the tool's endpoint. Both model architectures support tools. Null or an empty list names none.
- voicesType: array of AgentVoice …32nullableProperties: 2
The voices this Call Agent's calls may speak in, each a provider and voice id taken from GET /v1/available-voices. Each call this agent places picks one entry from the list at random, so a list of one always uses that voice and every entry in a longer list has an equal chance on each call. At most 32 entries. Null or an empty list keeps the voice each deployment is configured with — the behavior every agent had before this field existed. To keep a persona name and its voice consistent, give the name one voice, or several voices of one gender; read each voice's gender from GET /v1/available-voices. Resia does not enforce this — a name and a voice of any gender are allowed.

