Synthflow AI
AIRun AI voice phone agents with Synthflow AI. Agents create and configure assistants, attach phone numbers, place outbound calls, and pull back call status, transcripts, recordings, and collected variables.
What This Integration Enables
The phone is the highest-stakes channel a company automates. A bad email campaign embarrasses you in an inbox; a bad voice agent holds a thousand flawed conversations out loud, in your company's voice, with people who did not expect a robot. That is exactly why voice agents need orchestration more than any other AI capability. Synthflow provides the voice: no-code assistants with configurable prompts, voices, and phone numbers. FlowRunner provides everything around the voice: the context that personalizes each call, the gate that decides whether dialing is justified, and the pipeline that turns transcripts back into structured operational data. - Create, inspect, update, and retire voice assistants, with prompt and voice changes applying only to future calls - Attach workspace phone numbers to inbound agents and choose caller IDs for outbound - Place outbound calls with per-call custom variables injected into the agent's prompt - Poll completed calls for status, transcript, recording, duration, end-call reason, and collected variables The connector's own catalog is blunt about the stakes: Start Outbound Call places a real, billable call the moment it fires. FlowRunner treats that line as architecture. Everything before the dial tone is preparation an agent can do alone; the dial tone itself sits behind a [human-in-the-loop](/concepts/human-in-the-loop) gate.
Without FlowRunner
With FlowRunner
Use Case Scenarios
The qualification line that never sleeps, but always asks
Leads from the website flow into [HubSpot](/integrations/hubspot). A FlowRunner agent assembles each batch with phone numbers, consent status, and context, and holds it for the sales manager's approval. On approval, Start Outbound Call works through the list, each call personalized by custom variables. As Get Call returns completed transcripts, the agent writes budget, timeline, and interest level back to the CRM and posts the promising ones to [Slack](/integrations/slack). Reps spend their day talking to people who already said yes to a conversation.
Reminder calls with a decision tree
The evening before each appointment, the reminder flow dials confirmed bookings with the time, location, and provider name passed as variables. The voice agent collects a confirm, cancel, or reschedule intent as a variable, and FlowRunner routes accordingly: confirmations log quietly, cancellations release the slot, reschedule requests open a task for the front desk with the transcript attached. The calls nobody had time to make now happen every evening, and the calendar reflects reality by morning.
Standing up an inbound line without a switchboard project
Create Assistant builds an after-hours intake agent, List Voices picks the voice, and List Phone Numbers finds an unattached workspace number to route to it. Overnight callers reach a patient, consistent intake conversation instead of voicemail. Each morning, List Calls pulls the night's history into [Google Sheets](/integrations/google-sheets) with durations and end-call reasons, and an operations lead reviews the transcripts before any prompt change goes back through Update Assistant.
Human-in-Loop Highlight
Start Outbound Call is the spend that cannot be recalled in its purest form: it makes a real phone ring in a real person's pocket, and telephony plus LLM minutes bill the moment the call connects. There is no unsending a phone call. So in FlowRunner, no agent dials from its own judgment. Call lists are assembled by machine and approved by a named person, with the numbers, the calling window, and the assistant's current prompt in front of them. Update Assistant gets the same respect for a quieter reason: a prompt edit rewrites what the voice says to every future caller, so it ships like customer-facing copy, reviewed before it speaks. And Delete Assistant is gated because it releases the agent's phone number and can orphan the historical call logs tied to its model ID. The transcripts flow at machine speed. The decision to put your company's voice on someone's phone belongs to a human, every time.
Agent Capabilities
10 actionsAssistants
5- List Assistants Retrieves the voice agents configured in the workspace, each with its prompt, greeting, voice, language, LLM, and attached phone number. Paginated with limit and offset. The discovery step for the model_id every other operation needs.
- Get Assistant Retrieves a single agent's complete configuration by model ID: system prompt, greeting, voice settings, language, LLM selection, recording flag, and webhook URLs. The inspection step before updates and before placing calls.
- Create Assistant Creates a new no-code voice agent. The agent type determines its use: inbound agents answer an attached number, outbound agents place calls, widget agents run in the browser. Creating an agent places no calls and is not billed per minute.
- Update Assistant Updates an existing agent. Only supplied fields change, and changes apply solely to calls started after the update. Prompt edits ship through review, because they rewrite what the voice says.
- Delete Assistant Permanently deletes a voice agent, releasing its attached phone number and potentially orphaning its historical call logs. Cannot be undone, and never runs without a named approver.
Calls
3- Start Outbound Call Places a real, billable outbound call: the selected agent dials the recipient and holds a live conversation, consuming telephony and LLM minutes once connected. Queued asynchronously, returning a call_id to poll. Custom variables personalize each call through the agent's prompt. Behind this page's human gate, always.
- List Calls Retrieves an agent's call history with transcript, recording URL, duration, end-call reason, and collected variables, filterable by date range, status, duration window, and the lead's phone number.
- Get Call Retrieves a single call by ID with final status, full transcript, recording URL, duration, both phone numbers, end-call reason, and collected variables. Transcript and recording appear once the call ends.
Phone Numbers
1- List Phone Numbers Retrieves the telephony numbers provisioned in the workspace, each with its Twilio SID and availability for attachment to an inbound agent. Requires the Workspace ID configuration.
Voices
1- List Voices Retrieves the synthetic voices available to the workspace with voice_id, display name, gender, supported languages, and preview URL, with name search and provider filtering. The returned voice_id feeds Create Assistant and Update Assistant.
Frequently Asked Questions
What can FlowRunner do with Synthflow AI?
FlowRunner agents can run List Assistants, Get Assistant, and Create Assistant in Synthflow AI, plus 7 more actions.
Does connecting Synthflow AI to FlowRunner require OAuth?
No. Synthflow AI connects to FlowRunner with an API key, no OAuth flow required.
Can Synthflow AI trigger a FlowRunner workflow automatically?
Synthflow AI doesn't currently expose triggers in FlowRunner. It connects as an action step inside workflows started by another trigger.
Start building with Synthflow AI
$100 in credits. No card required. Connect in minutes.