Fish Audio connector for AI agents

Tools your voice ops agent reaches for on Fish Audio, scoped per user.

CALL ANY TOOL

34 tools for voice agents in Fish Audio: create, configure, and publish voice agents, place outbound calls, read sessions and recordings, manage agent tools, knowledge sources, and phone numbers, and design and search voice models.

fishaudio_create_agent

Create agent

Create a voice agent as a draft, optionally with its initial configuration. Returns the new agent with agent_id, name, status and publication_state draft. Use this to start a new agent. Publish it with fishaudio_publish_agent before it can take calls; change settings later with fishaudio_update_agent_config.

Parameters

Name

Type

Required

Description

name

string

Required

Agent name, 1 to 120 characters. Example: Support Line Agent

config

object

Optional

Optional initial draft configuration, applied on top of the platform defaults. Configuration sections, each an object: prompt (system_prompt, first_message_mode off/fixed/prompt, first_message, first_message_prompt), voice (voice_id from a voice model, speaking_language, expressive, speed 0.5-2, tts_model s2.1-pro or drama-3), asr (speech recognition model, multilingual, keyterms), conversation (turn-taking, interruptions, max duration, recording, transfers), tools (tool_ids to attach, system tools), knowledge_base (knowledge_source_ids), llm (model), variables (default placeholder values), guardrails and analysis. Example: {"prompt": {"system_prompt": "You are a friendly support agent.", "first_message_mode": "fixed", "first_message": "Hi, how can I help?"}, "voice": {"voice_id": "802e3bc2b27e49c2995d23ef70e6ac89", "speaking_language": "en"}}

description

string

Optional

Optional description, up to 2000 characters. Example: Answers billing questions for inbound callers.

fishaudio_create_agent_tool

Create agent tool

Create a workspace tool that voice agents can call during conversations: a webhook to your HTTP endpoint, or a client tool your frontend runs. Returns the new tool with its ID and definition. Use this to give agents an action. Attach it with the tools.tool_ids field of fishaudio_update_agent_config, then fishaudio_publish_agent.

Parameters

Name

Type

Required

Description

name

string

Required

Tool name the agent sees, 1 to 120 characters. Example: lookup_order

arguments

array

Optional

Arguments the agent fills in, each with a name (up to 64 characters) and description. Example: [{"name": "order_id", "description": "The customer's order ID"}]

body_template

string

Optional

Request body for non-GET calls; {{argument}}, {{system.*}} and session variable placeholders are filled in at call time. Example: {"order_id": "{{order_id}}"}

content_type

string

Optional

Content type of the request body. Defaults to application/json.

description

string

Optional

What the tool does and when the agent should call it, up to 2000 characters. Example: Look up an order's status by order ID.

error_handling

string

Optional

passthrough (default) lets the agent see endpoint errors; hide hides them. One of: `passthrough`, `hide`.

execution_mode

string

Optional

blocking (default) waits for the result, fire_and_forget does not wait, background runs while the conversation continues. One of: `blocking`, `fire_and_forget`, `background`.

expects_response

boolean

Optional

Whether the agent waits for and uses the endpoint's response. Defaults to true.

headers

array

Optional

Request headers, each with name, value and kind (custom, authorization_bearer or authorization_basic). Credential values are write-only. Example: [{"name": "Authorization", "value": "secret-token", "kind": "authorization_bearer"}]

method

string

Optional

HTTP method for webhook tools. Defaults to POST. One of: `GET`, `POST`, `PUT`, `PATCH`, `DELETE`.

timeout_seconds

integer

Optional

Seconds to wait for your endpoint. Defaults to 30.

tool_type

string

Optional

webhook (default) calls your HTTP endpoint during the conversation; client is run by your own frontend through the SDK and has no URL. One of: `webhook`, `client`.

url

string

Optional

Endpoint URL for webhook tools; may contain {{argument}} placeholders. Example: https://api.example.com/orders/{{order_id}}

fishaudio_create_phone_call

Call create phone

fishaudio_delete_agent

Delete agent

fishaudio_delete_agent_tool

Delete agent tool

fishaudio_delete_knowledge_source

Delete knowledge source

fishaudio_delete_voice_model

Delete voice model

fishaudio_design_voice

Design voice

fishaudio_end_agent_session

End agent session

fishaudio_get_agent

Get agent

fishaudio_get_agent_config

Get agent config

fishaudio_get_agent_session

Get agent session

fishaudio_get_agent_session_recording

Get agent session recording

fishaudio_get_agent_tool

Get agent tool

fishaudio_get_api_credit

Get api credit

fishaudio_get_knowledge_source

Get knowledge source

fishaudio_get_package

Get package

fishaudio_get_phone_number

Get phone number

fishaudio_list_agent_sessions

List agent sessions

List voice agent sessions (calls and web conversations), newest first, with optional filters. Returns sessions with session_id, agent_id, agent_name, status, end_reason, direction, caller_number, dialed_number, created_at, duration_seconds and metadata, plus has_more and next_cursor. Use this to find calls. Use fishaudio_get_agent_session for one session's transcript and tool calls.

Parameters

Name

Type

Required

Description

agent_id

string

Optional

ID of the agent whose sessions to return, from fishaudio_list_agents or fishaudio_create_agent (the agent_id field). Example: 4f2a8c6e1b3d5f7a9c0e2d4b6a8f1c3e

call_sid

string

Optional

Exact carrier call ID, such as a Twilio CallSid on inbound calls.

caller_number

string

Optional

Exact caller phone number in E.164 format (phone sessions only). Example: +14155550123

created_after

string

Optional

Only sessions created after this ISO 8601 time. Example: 2026-10-01T00:00:00Z

created_before

string

Optional

Only sessions created before this ISO 8601 time. Example: 2026-10-20T00:00:00Z

cursor

string

Optional

Cursor for the next page: pass the next_cursor value from the previous response while has_more is true. Do not combine with page.

direction

string

Optional

Call direction: inbound or outbound. One of: `inbound`, `outbound`.

end_reason

string

Optional

Comma-separated end reasons to include, such as user_hangup, agent_hangup, escalated, conversation_timeout, internal_error, cancelled, dial_failed. Example: user_hangup,agent_hangup

end_user_id

string

Optional

Exact end_user_id set when the session was created. Example: user-42

include_total

boolean

Optional

Set to true to also return the total count when paging with cursor. Defaults to false.

page

integer

Optional

1-based page number for offset pagination; the response then includes total. Do not combine with cursor. Example: 2

page_size

integer

Optional

Number of results per page, 1 to 100. Defaults to 30.

sip_call_id

string

Optional

Exact SIP Call-ID of the call (phone sessions only), to match a carrier record.

status

string

Optional

Comma-separated statuses to include: pending, active, completed, failed, unknown. Example: completed,failed

fishaudio_list_agent_tools

List agent tools

fishaudio_list_agent_versions

List agent versions

fishaudio_list_agents

List agents

fishaudio_list_knowledge_sources

List knowledge sources

fishaudio_list_my_voice_models

List my voice models

fishaudio_publish_agent

Publish agent

fishaudio_search_voice_models

Search voice models

fishaudio_update_agent

Update agent

fishaudio_update_agent_config

Update agent config

Patch a voice agent's draft configuration section by section; sections you leave out keep their value. Returns the full updated draft configuration with its new config_hash. Use this to change the agent's voice, prompt, speech recognition, conversation behavior, tools or knowledge. Changes reach live calls only after fishaudio_publish_agent.

Parameters

Name

Type

Required

Description

agent_id

string

Required

ID of the agent, from fishaudio_list_agents or fishaudio_create_agent (the agent_id field). Example: 4f2a8c6e1b3d5f7a9c0e2d4b6a8f1c3e

analysis

object

Optional

Post-call analysis section: summary, data_fields (values to extract from each call) and criteria (success criteria to score). Example: {"summary": {"enabled": true}}

asr

object

Optional

Speech recognition section: model (deepgram:nova-3, elevenlabs:scribe_v2_realtime, elevenlabs:scribe_v2_medical or fish_audio:transcribe-stream-1), multilingual, strict_language, keyterms (up to 50 words to recognize). Example: {"model": "deepgram:nova-3", "keyterms": ["Acme", "refund"]}

conversation

object

Optional

Conversation section: turn-taking and call handling, for example max_duration_seconds, response_wait_ms, eagerness (relaxed, balanced, eager), interruptible, interruption_sensitivity (low, balanced, high), hangup_after_seconds, record_audio, timezone, transfer_destinations. Example: {"max_duration_seconds": 900, "eagerness": "balanced", "record_audio": true}

guardrails

object

Optional

Guardrails section: focus (keep the agent on topic) and manipulation (resist prompt manipulation), both true/false. Example: {"focus": true, "manipulation": true}

knowledge_base

object

Optional

Knowledge base section: enabled and knowledge_source_ids (replaces the attached sources; IDs from fishaudio_list_knowledge_sources). Example: {"enabled": true, "knowledge_source_ids": ["3e5a7c9b1d2f4e6a8c0b2d4f6a8c1e3b"]}

llm

object

Optional

Language model section: model, a platform model ID such as anthropic/claude-sonnet-4.6 or openai/gpt-4.1 (the vendor's current list is in the API reference). Example: {"model": "anthropic/claude-sonnet-4.6"}

prompt

object

Optional

Prompt section: system_prompt (up to 32000 tokens; under 2000 recommended), first_message_mode (off, fixed or prompt), first_message, first_message_prompt. Example: {"system_prompt": "You are a friendly support agent for Acme.", "first_message_mode": "fixed", "first_message": "Hi, thanks for calling Acme!"}

tools

object

Optional

Tools section: enabled, tool_ids (replaces the attached tool set; IDs from fishaudio_list_agent_tools) and system_tools. Example: {"enabled": true, "tool_ids": ["tool_8f2c4e6a1b3d"]}

variables

object

Optional

Variables section: defaults (values for {{placeholders}} in the prompt when a session supplies none; the map is replaced as a whole) and on_missing (keep, empty or reject). Example: {"defaults": {"company": "Acme"}, "on_missing": "empty"}

voice

object

Optional

Voice section: voice_id (any public voice model ID), speaking_language (ISO 639-1 code such as en), expressive (true/false), speed (0.5 to 2), tts_model (s2.1-pro or drama-3), pronunciation_dictionary_ids. Example: {"voice_id": "802e3bc2b27e49c2995d23ef70e6ac89", "speaking_language": "en", "speed": 1.1}

fishaudio_update_agent_tool

Update agent tool

Update fields of a voice agent tool; fields you leave out keep their value. Returns the updated tool. Attached agents' drafts pick up the change at once; published versions change only after re-publishing. Use this to change a tool's endpoint, arguments or headers. Use fishaudio_get_agent_tool to read it first.

Parameters

Name

Type

Required

Description

tool_id

string

Required

Agent tool ID, from fishaudio_list_agent_tools. Example: tool_8f2c4e6a1b3d

arguments

array

Optional

Arguments the agent fills in, each with a name (up to 64 characters) and description. Example: [{"name": "order_id", "description": "The customer's order ID"}]

body_template

string

Optional

Request body for non-GET calls; {{argument}}, {{system.*}} and session variable placeholders are filled in at call time. Example: {"order_id": "{{order_id}}"}

content_type

string

Optional

Content type of the request body. Defaults to application/json.

description

string

Optional

What the tool does and when the agent should call it, up to 2000 characters. Example: Look up an order's status by order ID.

error_handling

string

Optional

passthrough (default) lets the agent see endpoint errors; hide hides them. One of: `passthrough`, `hide`.

execution_mode

string

Optional

blocking (default) waits for the result, fire_and_forget does not wait, background runs while the conversation continues. One of: `blocking`, `fire_and_forget`, `background`.

expects_response

boolean

Optional

Whether the agent waits for and uses the endpoint's response. Defaults to true.

headers

array

Optional

Request headers, each with name, value and kind (custom, authorization_bearer or authorization_basic). Credential values are write-only. Sending this replaces the whole list, so include credential values again. Example: [{"name": "Authorization", "value": "secret-token", "kind": "authorization_bearer"}]

method

string

Optional

HTTP method for webhook tools. Defaults to POST. One of: `GET`, `POST`, `PUT`, `PATCH`, `DELETE`.

name

string

Optional

Tool name the agent sees, 1 to 120 characters. Example: lookup_order

timeout_seconds

integer

Optional

Seconds to wait for your endpoint. Defaults to 30.

url

string

Optional

Endpoint URL for webhook tools; may contain {{argument}} placeholders. Example: https://api.example.com/orders/{{order_id}}

fishaudio_update_phone_number

Update phone number

For more tools, view docs.

Build your Agent

Same auth pattern across LangChain, OpenAI, Anthropic, and Google ADK.

import { ScalekitClient } from "@scalekit-sdk/node";
import { createReactAgent } from "@langchain/langgraph/prebuilt";

const sk = new ScalekitClient(env.SCALEKIT_ENV_URL, env.SCALEKIT_CLIENT_ID, env.SCALEKIT_CLIENT_SECRET);

// Fish Audio tools scoped to this user
const { tools } = await sk.tools.listScopedTools("user_123", {
  filter: { connectionNames: ["fishaudio"], toolNames: [
    "fishaudio_list_agents",
    "fishaudio_create_phone_call",
    "fishaudio_get_agent_session"] },
  pageSize: 100,
});

const agent = createReactAgent({ llm, tools });
await agent.invoke({ messages: [{ role: "user", content: "Call the customer with the booking agent and show the transcript" }] });

import OpenAI from "openai";
import { ScalekitClient } from "@scalekit-sdk/node";

const sk = new ScalekitClient(env.SCALEKIT_ENV_URL, env.SCALEKIT_CLIENT_ID, env.SCALEKIT_CLIENT_SECRET);
const openai = new OpenAI();

const { tools } = await sk.tools.listScopedTools("user_123", {
  filter: { connectionNames: ["fishaudio"] }, pageSize: 100,
});

const res = await openai.chat.completions.create({
  model: "gpt-5",
  messages: [{ role: "user", content: "Call the customer with the booking agent and show the transcript" }],
  tools,
});

// Execute the tool call with the user's vaulted Fish Audio access token
await sk.tools.executeTool(res.choices[0].message.tool_calls[0], "user_123");

import Anthropic from "@anthropic-ai/sdk";
import { ScalekitClient } from "@scalekit-sdk/node";

const sk = new ScalekitClient(env.SCALEKIT_ENV_URL, env.SCALEKIT_CLIENT_ID, env.SCALEKIT_CLIENT_SECRET);
const anthropic = new Anthropic();

const { tools } = await sk.tools.listScopedTools("user_123", {
  filter: { connectionNames: ["fishaudio"] }, pageSize: 100,
});

const msg = await anthropic.messages.create({
  model: "claude-sonnet-5",
  max_tokens: 1024,
  messages: [{ role: "user", content: "Call the customer with the booking agent and show the transcript" }],
  tools,
});

// Tool call runs with the user's vaulted Fish Audio access token
await sk.tools.executeTool(msg.content, "user_123");

import { Agent } from "@google/adk/agents";
import { ScalekitClient } from "@scalekit-sdk/node";

const sk = new ScalekitClient(env.SCALEKIT_ENV_URL, env.SCALEKIT_CLIENT_ID, env.SCALEKIT_CLIENT_SECRET);

const { tools } = await sk.tools.listScopedTools("user_123", {
  filter: { connectionNames: ["fishaudio"] }, pageSize: 100,
});

const agent = new Agent({
  name: "voice_ops_agent",
  model: "gemini-2.5-pro",
  instruction: "Fish Audio tools scoped to this user",
  tools,
});

await agent.run("Call the customer with the booking agent and show the transcript");

Try these prompts

Copy any prompt into your agent. Each maps directly to a Fish Audio tool. Click to copy, paste into your agent, done.

Voice agents

Copy the prompt

Copied

Create a draft voice agent for appointment booking.

Copy the prompt

Copied

Publish the booking agent's current draft as a new version.

Copy the prompt

Copied

Show the published versions of the support agent.

Calls and sessions

Copy the prompt

Copied

Call +1 415 555 0199 with the booking agent.

Copy the prompt

Copied

Show the transcript of yesterday's longest session.

Copy the prompt

Copied

Get the recording links for the last session.

Voices and account

Copy the prompt

Copied

Design a calm, friendly voice for our receptionist agent.

Copy the prompt

Copied

Search the public voice library for British English voices.

Copy the prompt

Copied

How much credit is left on our Fish Audio account?

SEE HOW AUTH WORKS

Each user connects their own Fish Audio access token once, and Scalekit sends it with every call. Access tokens stay vaulted, every call is scope checked, and every action is logged.

Authorize

Your user connects

Fish Audio

once. We tie it to their identity and the meetings they approved — no shared bot account, no org-wide access

Who:

user ‘A’

when:

Once per user

access:

Limited to user

Store

Their

Fish Audio

token lives in a vault scoped to them. User A's meetings are never reachable by an agent acting for user B, even on the same connection

vault:

encrypted

scope:

per-user

tokens:

auto-refreshed

Resolve

When your agent calls a

Fish Audio

tool, we fetch the right token server-side. It never touches your agent, never appears in the LLM context, never shows up in your logs

speed:

~40ms

check:

before every call

seen by:

nobody

Audit

Every

Fish Audio

tool call is logged — who triggered it, which meeting was fetched, what came back. 90 days of history, tied to the user who authorized it

history:

90 days

export:

SIEM-ready

logged:

every call

Test other agents

See the same per-user auth pattern across other AI and media connectors.

Support and Ops Teams

Support ticket automation agent

Fetches new Zendesk tickets, drafts a reply from Notion knowledge base articles, digests what it cannot answer to Slack, and archives the rest, acting as the support agent rather than a shared API key.

GTM and RevOps Teams

Outbound prospecting agent

Searches Apollo for prospects matching your ICP, scores and ranks them, drafts personalized outreach in Gmail, and logs every send to Google Sheets. Mail goes out as the rep, not from a shared inbox.

Why Scalekit

Secure your agent's access. Connectors ship in minutes

01.

Shared tokens break per-user analytics

A shared Fish Audio access token looks fine in a demo. In production every outbound call and agent publish looks like one service account, and you cannot tell which user triggered it. Scalekit resolves the credential of the actual user who triggered the agent, never a shared bot.

// shared key
audit → bot_service_account

// scalekit
audit → user_abc ✓

02.

Authentication is not authorization

03.

Multi-tenancy is architectural

04.

Fish Audio today. Ten connectors tomorrow.

“Our agents act across Salesforce, Gong, Google Drive, and more, on behalf of every customer. Scalekit behind the scenes meant we can keep adding tools without ever rebuilding how credentials or tool calling work.”

Venu Madhav Kattagoni

Head of Engineering / Von

FAQs

Frequently Asked Questions

Does the agent access Fish Audio as the user or through a shared access token?

As the user. Each user connects their own Fish Audio access token once, and Scalekit sends it with every call. Audit logs attribute every call and agent change to that user, not a shared service account.

Where is the Fish Audio access token stored?

In Scalekit's managed AES-256 token vault, namespaced per tenant. To rotate it, the user reconnects with a new access token. Revocation is a single dashboard action. Credentials never appear in prompts, logs, or LLM context.

Can I limit what the agent does in Fish Audio?

Yes. Filter by tool name in listScopedTools to expose only what you want, for example sessions, transcripts, and recordings without fishaudio_create_phone_call or fishaudio_delete_agent, which permanently deletes an agent with its version history. Scalekit also enforces scope checks before every API call.

What happens when a user revokes Fish Audio access?

The connection is invalidated on the next tool call. Subsequent requests for that user fail closed with a clear error. Other users in the tenant remain unaffected. The event is logged for audit.

Do changes to a voice agent go live right away?

No. Configuration changes edit the agent's draft. fishaudio_publish_agent freezes the draft as a new immutable version and makes it live for new sessions, and fishaudio_list_agent_versions shows the published history.