For AI agents: a documentation index is available at the root level at /llms.txt. Append /llms.txt to any URL for a page-level index, or .md for the markdown version of any page.
Generates one or more alternative assistant responses for a conversation
you supply inline to simulate Phonic agent behavior.
This endpoint is stateless, so it does not create a new conversation or
store anything. The request carries the system prompt, a conversation so
far as `input`, and the tools the assistant may call.
There are two different types of tool configurations:
- `tool_definitions`: defined inline. Typically if you use LiveKit, this
is where you can pass in your tool definitions.
- `tools`: predefined tools configured in Phonic, referenced by name, in
the same shape as the STS WebSocket `config` message. Today this covers
built-in tools such as `choose_not_to_respond` and transfer tools stored
in `project`. Your context, webhook, MCP and WebSocket tools cannot be
referenced here yet; define those inline as `tool_definitions`.
Each item in `input` is a user message, an assistant message (with
optional `tool_calls` or an `action`), or a `tool_call_output`. Every
assistant tool call must be followed immediately by the
`tool_call_output` item that carries its result.
A tool referenced in `tools` resolves to an `action` on the generated
response rather than a tool call - the event a live conversation would
have emitted, without the effect actually being carried out.
This is an experimental feature and must be enabled for your workspace;
otherwise, it returns `404`. Please contact our team if you would like
access.
Authentication
AuthorizationBearer
Bearer authentication header of the form `Bearer <PHONIC_API_KEY>`. Manage your API keys [here](https://phonic.co/api-keys).
Request
This endpoint expects an object.
system_promptstringRequired>=1 character
The system prompt the assistant should follow.
inputlist of objectsRequired
The conversation so far, in order. Must contain at least one item.
voice_idstringOptionalDefaults to sabrina
ID of the voice the assistant would speak with. It shapes how the responses are worded.
default_languageenumOptional
ISO 639-1 language code that sets the assistant’s default language to recognize and speak.
additional_languageslist of enumsOptionalDefaults to []
Array of additional ISO 639-1 language codes that the assistant should be able to recognize and speak. Should not include default_language.
tool_definitionslist of objectsOptionalDefaults to []
Tools defined inline for this request only. Typically if you use LiveKit, this is where you can pass in your tool definitions. Names must be unique, must not repeat a name in tools, and cannot be one of the names Phonic reserves for its built-in tools.
projectstringOptional>=1 character
Name of the project the tools referenced by name in tools belong to. Required whenever tools names a tool stored in your workspace; built-in tools can be referenced without it.
toolslist of strings or objectsOptionalDefaults to []
Tools the assistant may call that already exist - a built-in tool, or a transfer tool stored in project, referenced by name. Names must be unique and must not repeat a name in tool_definitions. Stored tools that are not transfer tools cannot be referenced here yet; define them inline as tool_definitions instead.
phonic_modelenumOptional
The Phonic speech-to-speech model to generate with. Omit it to use the current default model.
Allowed values:
num_responsesintegerOptional1-200Defaults to 1
Number of alternative responses to generate.
Response
Success response
responseslist of objects
The generated responses - always num_responses of them.
Errors
400
Bad Request Error
401
Unauthorized Error
404
Not Found Error
429
Too Many Requests Error
500
Internal Server Error
502
Bad Gateway Error
503
Service Unavailable Error
Generates one or more alternative assistant responses for a conversation
you supply inline to simulate Phonic agent behavior.
This endpoint is stateless, so it does not create a new conversation or
store anything. The request carries the system prompt, a conversation so
far as input, and the tools the assistant may call.
There are two different types of tool configurations:
tool_definitions: defined inline. Typically if you use LiveKit, this
is where you can pass in your tool definitions.
tools: predefined tools configured in Phonic, referenced by name, in
the same shape as the STS WebSocket config message. Today this covers
built-in tools such as choose_not_to_respond and transfer tools stored
in project. Your context, webhook, MCP and WebSocket tools cannot be
referenced here yet; define those inline as tool_definitions.
Each item in input is a user message, an assistant message (with
optional tool_calls or an action), or a tool_call_output. Every
assistant tool call must be followed immediately by the
tool_call_output item that carries its result.
A tool referenced in tools resolves to an action on the generated
response rather than a tool call - the event a live conversation would
have emitted, without the effect actually being carried out.
This is an experimental feature and must be enabled for your workspace;
otherwise, it returns 404. Please contact our team if you would like
access.
Bearer authentication header of the form Bearer <PHONIC_API_KEY>. Manage your API keys here.
