Synthflow

What it is

Synthflow is a no-code platform for building AI phone agents, and since 2026 SMS and WhatsApp chat agents too, aimed at contact centers and the BPOs and resellers that serve them. An operations or CX person builds the agent in a visual editor, connects the calendar, CRM and phone system, tests it against simulated callers, and publishes, without writing a webhook. That puts it in a different seat from Retell AI and Vapi, which assume an engineer owns the agent, and from PolyAI or Parloa, which build the agent for you.

The company was founded in Berlin in 2023 by CEO Hakob Astabatsyan, CPO Albert Astabatsyan and CTO Sassun Mirzakhan-Saky. It has raised $30M in total, including a $20M Series A led by Accel in June 2025, and lists about 70 employees. Synthflow says it handles more than 1 million calls a day; its April 2026 press release put the cumulative figure at over 65 million calls for more than 100 enterprise customers, including Freshworks and Thryv. Both are vendor numbers.

Why it shows up in CX and support stacks

• It plugs into the contact center you already run. A Five9 handoff action (July 2026), an 8x8 Contact Center integration (August 2026), Zoom Phone and Zoom Contact Center (June 2026), and RingCX over SIP let the agent answer calls inside the existing platform and pass the caller to a human queue with context. Batteries Plus runs it on Zoom Phone across more than 700 US locations.
• Channel partners sell and deploy it. 8x8 signed a strategic partnership in April 2026 that allows 8x8 and its partners to resell Synthflow; AVANT Communications signed an exclusive distribution deal on 23 June 2026. White-label is included in enterprise contracts, which is why BPOs run it under their own brand.
• One agent across voice and messaging. Agents can send SMS and WhatsApp messages mid-call (April 2026), and memory has been shared across voice, SMS and WhatsApp since June 2026, so a caller who gets a confirmation text can reply and reach the same agent with context.
• Testing a non-engineer can run. Simulations and test suites run from the agent editor (September 2026), and any two agent versions can be compared side by side (October 2026), which is the guard against a prompt edit quietly breaking a booking flow.
• MCP both ways. MCP Connectors (May 2026) let a live agent call tools on external MCP servers, and the Synthflow MCP server ships as an official Claude connector (July 2026) and ChatGPT plugin (September 2026), so you can inspect and edit agents from those clients.
• Compliance paperwork exists. The trust center lists SOC 2 Type 2, ISO 27001, GDPR, PCI DSS v4.0.1 and HIPAA (self-attested), and an EU hosting region opened in May 2026.

Pricing reality

Synthflow is now enterprise-only. Self-serve plans were replaced by pay-as-you-go in June 2026, then pay-as-you-go was removed the same month: a pricing tracker captured $0.09 per minute for the voice engine plus LLM cost from $0.02 per minute on 9 June, and on 24 June a single Enterprise plan with “contracts start at $30,000 annually.” The live pricing page no longer shows any figure, has no free plan or trial, and sells an annual agreement scoped to call volume, phone system, integrations and launch support by forward-deployed engineers.

Budget for at least $30K a year and treat it as a floor, not a quote. At the old $0.09 voice-engine rate alone, $30K covered about 330,000 minutes (about 270,000 once the cheapest LLM is added), so a single after-hours line handling 5,000 minutes a month pays well above what the same minutes would cost on Retell or ElevenLabs. Third-party pricing guides report overage near $0.45 to $0.50 per minute once included usage runs out; Synthflow does not publish an overage rate, so get one in writing.

Best for

CX and contact-center operations leaders, and BPOs, who need voice agents on Five9, 8x8, Zoom or RingCentral lines, where the person building and tuning the agent is an ops manager rather than an engineer, and the volume justifies a $30K-plus annual contract. Typical uses are appointment booking, order status, inbound qualification and after-hours overflow.

Skip it if you want to start on a credit card, if an engineer will own the agent anyway, or if you need to pick and swap the LLM yourself.

Versus the alternatives

Retell AI and Vapi are the two default voice-agent platforms. Retell bills one meter, near $0.15 per minute for a mainstream build, with QA built in; Vapi charges $0.05 per minute for orchestration and lets you choose every component. Both start pay-as-you-go. Pick either when you have an engineer and want to pilot without a contract; pick Synthflow when the builder is not an engineer or the agent must live inside Five9, 8x8 or Zoom.

ElevenLabs Agents is the fastest-growing entrant, at $0.08 per minute plus LLM on published plans; pick it when voice quality and language coverage decide the project. If you want a vendor to design and run the agent, buy managed: PolyAI or Parloa. If you run Five9 and its native AI already clears your containment bar, buy nothing extra.

If none fit, pilot your highest-volume intent on Retell’s or ElevenLabs’ self-serve tier first; that gives you the minute and containment numbers to negotiate a Synthflow contract with.

Watch-outs

• Packaging changed three times in 2026 alone. Flat tiers gave way to pay-as-you-go, pay-as-you-go to enterprise-only, and Stripe-based reselling for partners ended on 15 September 2026. Guard: fix the per-minute rate, overage rate and renewal cap for the full term in the order form, and keep agent configurations exported through the API.
• You don’t always control the model. Since August 2026, legacy non-enterprise accounts still on older self-serve plans run on Synthflow’s own LLM with GPT models removed, which shows the vendor will change the model under you. Guard: name the LLM, and the notice period for changing it, in the contract, and re-run your test suite on every model change.
• HIPAA is self-attested. There is no third-party HIPAA attestation in the trust center. Guard: require a signed BAA and the SOC 2 Type 2 report before any PHI touches a call, and keep sensitive intents on a human queue until both are in hand.
• Scale figures are vendor-reported and inconsistent. The newsroom says 3,000+ businesses and 1M+ calls a day; April 2026 press releases say 100+ enterprise customers and 65 million calls in total, which a 1M-a-day rate would reach in about two months. Guard: ask for two references at your volume on your phone system, and pilot with a measured containment target before signing the annual term.