Voiceflow
Also known as: Voiceflow AI
Collaborative platform for designing, building, and deploying AI agents across chat and voice.
Voiceflow is a collaborative platform for designing, building, testing, launching, and managing AI agents across both chat and voice channels. Founded in 2019 (originally to help teams build Alexa Skills), it has evolved into an enterprise conversational AI platform used by more than 250,000 users and teams worldwide, with customers including Turo, StubHub International, and Trilogy. Customer experience is the platform's center of gravity: Voiceflow positions itself for enterprise CX teams, and its solution pages cover support automation, chatbots, answering services, call centers, appointment booking, and virtual receptionists.
At its core is a visual drag and drop canvas where teams map conversation flows from modular blocks rather than writing code, then reuse saved Components (such as an authentication or intake flow) across agents and channels. A Workflow Builder handles multi-step tasks, and a vector-based Knowledge Base lets agents answer from a company's own documents and data.
Voiceflow is model-agnostic (agents can be powered by any LLM provider, including OpenAI and Anthropic) and supports pro-code extension through custom JavaScript Functions, API blocks, and platform APIs that control agent state, knowledge base, transcripts, and analytics. Built-in simulation lets teams test agents against live conversations before shipping.
Collaboration is a defining feature: workspaces, versioning, commenting, and real-time co-editing make agent design a team activity across departments and countries. Finished agents deploy as embeddable web chat and voice widgets, into mobile apps and smart speakers, across telephony and call centers, over SMS, or through APIs to any interface.
Voice is a first class channel rather than an add on: the same agent runs across web chat, in app, WhatsApp, SMS, and phone, and Voiceflow integrates with contact center platforms including Genesys, NICE CXone, Five9, Amazon Connect, and Twilio Flex alongside helpdesks such as Zendesk, Salesforce Service Cloud, ServiceNow, Freshdesk, and Intercom. The platform is built for enterprise with SOC 2, ISO/IEC 27001:2022, and GDPR compliance, SSO, and granular permissions. Common uses span customer support deflection, voice call automation, in-product copilots, and lead qualification. Voiceflow does not publish plan prices: a free plan sits alongside Pro, Business, and Enterprise tiers, with plan fees, editor seats, and usage credits shown inside the product or quoted on request.
Vendor details
Canonical URL
https://www.voiceflow.com
Category
Customer support agent
Subcategory
Conversational AI agent design and build platform for chat and voice
Funding status
Independent, venture-backed conversational AI platform; founded in 2019 (originally for building Alexa Skills). 250,000+ users and teams worldwide.
Company status
independent
Use cases & customers
Primary use cases
Target customers
Deployment options
Integrations
Model-agnostic — power agents with any LLM provider, including OpenAI and Anthropic. Connects to CRMs, databases, webhooks, and business apps (Salesforce, Shopify, Zendesk, Snowflake) via API blocks and custom JavaScript Functions; deploy to website chat/voice widgets, mobile apps, smart speakers, telephony/call centers, SMS, and any interface through platform APIs. Enterprise security: SOC 2, ISO/IEC 27001:2022, GDPR, SSO, and secrets management.
In practice
Your team maps conversation flows in slides, then hands them to engineers to rebuild. Voiceflow's visual canvas lets non-engineers design, test, and ship the agent directly.
You're building separate bots for chat and for voice. Voiceflow designs both channels on one platform and deploys them as web widgets, into telephony, or over SMS.
Agent design gets stuck with one person while everyone else waits. Voiceflow adds workspaces, versioning, and real-time co-editing, so the whole team builds together.
Sources & related URLs
Agentic Index coverage score
9.0 / 14 capabilities · 64%
| Integrations & Tool Calling | Full |
|---|---|
|
Agents act in any third-party system through a build step, not a fixed connector list. Voiceflow says agents integrate with third-party services and APIs using Voiceflow Functions and API blocks, and that teams build custom logic, actions and integrations with the full power of code, APIs and JSON. Because the route is a general API block plus custom code, the reach is set by what the customer can call, and a separate integrations directory sits alongside. App Credentials are stored encrypted, and API keys never appear in agent logs. Sourcevoiceflow.com/features/platform-overviewread 2026-09-05 |
|
| Workflow Orchestration | Full |
|
Two orchestration modes sit side by side. Workflow Builder handles multi-step tasks as scripted flows the agent calls, which execute in order every time, and Voiceflow calls this the route for ultimate control. Playbooks carry natural-language instructions for complex tasks the agent executes, the reasoning-led alternative. Beneath both sits a set of canvas steps covering agent, prompt, capture, condition, set and end, so branching, variable assignment and termination are all built in, and reusable Components carry a sequence such as authentication across agents. Sourcevoiceflow.com/features/workflowsread 2026-09-05 |
|
| Knowledge Grounding & RAG | Full |
|
Customers curate a knowledge store the agents answer from. Knowledge Base is a named product with its own feature page. Teams import data and documents into it so agents respond using curated data, and Voiceflow describes it as training agents on the customer's own data to power responses and perform tasks using any LLM model. Voiceflow's build sequence puts the knowledge base first, before workflows and integrations, and the Content Management System curates agent content, components and logic so it scales across teams and use cases. Sourcevoiceflow.com/features/knowledge-baseread 2026-09-05 |
|
| Human Oversight & Guardrails | Not documented |
|
There is no runtime oversight of agent actions, with no handoff or escalation to a human agent, no approval step before sending and no policy threshold that pauses a sensitive action. The controls that exist sit around the run. Staging and stakeholder review before production is a release gate, SSO with advanced user permissions is an administrative control, and neither holds an agent action at run time. |
|
| Security, Identity & Governance | Full |
|
Voiceflow holds SOC 2 Type II, ISO/IEC 27001:2022, GDPR and HIPAA, each linked to a trust portal, alongside annual external penetration testing, continuous vulnerability scanning, automated and manual code review, and a vulnerability disclosure policy with safe harbor and a bug bounty. For customers, SSO with SAML and roles and permissions each have their own workspace administration sections, with granular access controls, audit logging and alerting, encryption at rest under customer-managed keys, and full project history with rollback. Infrastructure sits on AWS and GCP with multi-AZ databases, point-in-time recovery, TLS and mTLS in transit, and a CDN-backed WAF. That portal sits at trust.voiceflow.com. Sourcevoiceflow.com/legal/securityread 2026-09-05 |
|
| Observability & Auditability | Partial |
|
Transcripts and Analytics are their own sections under Monitoring Agents, audit logging and alerting are a platform property alongside full project history with rollback, and Environments and Backups give a record of what was deployed when. Voiceflow does not show whether a transcript reconstructs a run step by step, with the blocks traversed, the tool calls made and the retrieval performed. Sourcevoiceflow.com/legal/securityread 2026-09-05 |
|
| Memory & State Persistence | Not documented |
|
There is no memory, saved state or persistence layer. The Set step assigns variables within a flow, which is session state, and the Knowledge Base is a curated corpus for grounding answers, not memory the agent keeps. |
|
| Deployment & Data Residency | Partial |
|
A different hosting arrangement is available by negotiation, not as a published option. Voiceflow is based in Canada and may host, transfer and process data in Canada and other countries unless it expressly agrees otherwise, so a different arrangement can be had by contract, with enterprise negotiation as the gate. No region is named or selectable, and there is no VPC, single tenant or self hosted option. Voiceflow runs multi-AZ databases, cross-region backup and infrastructure on AWS and GCP, which is redundancy, not a location a customer chooses. Voiceflow keeps a trust portal at trust.voiceflow.com. Sourcevoiceflow.com/privacyread 2026-09-05 |
|
| Prebuilt Agents, Templates & Packs | Partial |
|
Packaged assets exist and are reused across agents, but the customer builds them. Reusable Components hold a sequence such as an authentication or intake flow and are reused across agents and channels, the Content Management System curates agent content, components and logic so they scale across teams and use cases, and projects and functions can be exported and imported between workspaces. Every one of those assets starts with the customer. There is no catalog of prebuilt agents or templates from Voiceflow to browse and put into service, and the industry and solution pages describe use cases, not units to adopt. Sourcevoiceflow.com/features/agent-cmsread 2026-09-05 |
|
| Triggers & Channel Coverage | Full |
|
Coverage spans text, voice and messaging, with the interface left open. Voiceflow ships a web chat widget and, through the Dialog API, deployment into any custom interface, with worked examples for Slack, Microsoft Teams, Discord, Telegram, Twilio SMS, WhatsApp and Shopify. Voiceflow treats voice as a first-class channel, not an add-on, with its own documentation covering voice settings, voice input, call recording, custom voice actions and SIP functionality, and it reports 50ms platform latency for telephony. Answering service, call center and virtual receptionist solutions cover inbound telephony. Sourcedocs.voiceflow.com/docs/faqread 2026-09-05 |
|
| Model Flexibility & Routing | Full |
|
The customer chooses the model. Voiceflow names Use any model as a platform property, so any AI vendor, including OpenAI and Anthropic, can power agent responses, logic and actions, and the Knowledge Base trains agents on the customer's data to power responses and perform tasks using any LLM model. Model choice sits with the customer at build time, and Voiceflow frames the whole platform as building agents with any models and integrations. Sourcevoiceflow.com/features/platform-overviewread 2026-09-05 |
|
| APIs, SDKs & MCP Extensibility | Full |
|
The platform can be called from outside. Voiceflow offers a Dialog API for launching agents into any custom interface, an API reference alongside the documentation and changelog, and a Voiceflow CLI with its own install and authentication guides under Developer Tools. Agent API keys are issued per agent from the agent settings, so a third party can authenticate against a specific agent, not only through the console. Worked examples built on those APIs cover Slack, Microsoft Teams, Discord, Telegram, Twilio SMS, WhatsApp, Shopify and Unity, and Voiceflow treats a developer platform for custom features, integrations and interfaces as a product pillar. Sourcedocs.voiceflow.com/docs/faqread 2026-09-05 |
|
| Testing, Debugging & Optimization | Partial |
|
Production pipelines let teams test agent builds in staging and share them with stakeholders for feedback before production, a Test Tool has its own interface section, prototypes can be shared for review, and Environments and Backups separate work in progress from what is live. Those steps form a quality gate that ends in a person's judgment. There is also an Evaluations section under Monitoring Agents, and Voiceflow does not show whether it produces a score, pass rate or judge verdict against a set of scenarios. Sourcevoiceflow.com/features/platform-overviewread 2026-09-05 |
|
| Browser & Computer Use | Not documented |
|
There is no browser control, hosted session, virtual desktop or computer use. Agents work through programmatic routes, reaching third-party systems through Voiceflow Functions and API blocks and reaching end users through the web chat widget, the Dialog API and telephony, and Voiceflow frames itself as a developer platform for custom integrations and interfaces. A native web search tool retrieves information and does not operate a screen. Sourcevoiceflow.com/features/platform-overviewread 2026-09-05 |
|
The Agentic Index coverage score grades every vendor Full, Partial or Not documented against the same 14 buyer facing capabilities, from public evidence only. Each capability links to how all vendors in the index score on it. How this evidence is graded
Product updates
The Voiceflow MCP server now serves version 3 by default, giving AI clients one small vf_* tool per operation with schemas and how to guides loaded on demand through vf_describe_operations, vf_skill_list and vf_skill_load. Connections can be narrowed with URL options for toolsets, read_only=true and a single project. Clients that rely on the older grouped voiceflow_* tools keep them by adding ?version=2.
Bears on: MCP / tool calling / API
View sourceV3 projects can now route user messages with Voiceflow Classify, an intent classifier that uses Voiceflow's own decision model to pick the intent that best matches each message. It is faster than classifying with a large language model, needs no training, and returns no match instead of guessing when nothing clearly fits. When it cannot classify a message, for example with more than 254 intents in scope, the agent falls back to the LLM method for that message.
Bears on: Agent capability
View sourceAny Voiceflow tool, whether a function, API, MCP or integration tool, can now be run on its own from the API, the Voiceflow MCP server or its copilot, returning the result without starting a conversation. Secrets the tool needs are passed by name rather than value.
Bears on: MCP / tool calling / API
View sourcePricing
Not public · free plan
plan fee + add ons (seats, phone numbers, concurrent calls) + credits
Included quota
Not published. Each paid plan includes a monthly credit allotment that resets at the start of each billing cycle, with the amount shown in the in product Plans and Billing tab. The Free plan grant is one time and does not renew.
What is public
The plan structure is public, the prices are not. Voiceflow publishes four plans: Free (limited features, one time credit grant that does not renew, explicitly not suitable for production), Pro (individual builders, access to all LLM models, higher agent limit than Free), Business (growing teams, LLM fallback models, priority support), and Enterprise (unlimited product usage, SSO, private cloud hosting, custom credit allotments). The pricing page carries no dollar figures at all and splits into an Agencies and Partners path (free trial, no credit card, usage based billing, multi client workspace management, white labeling) and a Businesses path (Book a demo, Request pricing). Current pricing for plans, add ons, and credit bundles is visible only in the Plans and Billing tab inside the product.
Billing mechanics
Billing is at the organization level and combines a monthly plan fee, add ons, and credit based usage. Credits meter agent actions such as generating AI responses and making phone calls, are shared across all workspaces in the organization, and reset at the start of each billing cycle on paid plans; the Free plan grant is one time and does not renew. Add ons are billed monthly alongside the plan: editor seats (every paid plan includes some free seats and more can be purchased, while viewers are free and unlimited), one Voiceflow phone number per workspace with additional numbers available, and a concurrent calls limit that can be raised, which applies to calls but not to chat or API sessions. Most billing actions are self serve from Plans and Billing, including comparing and changing plans, managing add ons, updating payment method, and downloading invoices. Enterprise is billed per contract through an account manager.
Cost watchouts
Editor seats, extra phone numbers, and additional concurrent call capacity are all priced add ons and none of their rates are published. Voice burns credits far faster than chat. Telephony is billed separately through an external carrier. Auto top ups prevent outages by buying credits automatically, which also removes the natural stopping point on a runaway bill. Because no plan price is published, total cost cannot be estimated before signup or a sales conversation.
Variable cost rationale
Credit consumption dominates cost and voice consumes credits far faster than chat. The plan fee itself is now unpublished, so no component of the bill is quotable in advance. Editor seats, extra phone numbers, and additional concurrent call capacity are separate add ons, each at an unpublished price, and external telephony (Twilio, Vonage, Telnyx) is billed separately.
Additional watchouts
Credit burn on voice, unpublished plan and add on pricing, compounding seat and phone number costs, Enterprise gated security features, and separate telephony charges.
Overage / add-ons
Included credits reset at the start of each billing cycle. When credits run low you can upgrade the plan, buy a credit bundle, or enable auto top ups, which purchase credits automatically at zero so agents are not interrupted mid month. Bundle and top up rates are not published.
Sales call required
Mixed (some tiers require a call)
Free / trial
Free plan · free trial, no credit card
Lowest paid plan
Pro (price not published)
Commercial notes
Founded 2019; more than 250,000 users and teams. Restructured to a credits-based system on 29 April 2025. Withdrew all public plan pricing between 1 Jul and 28 Aug 2026, moving from a published tier table to an in-product and request-only model. Best fit for teams designing chat and voice agents; voice-heavy workloads escalate cost quickly.
Key ambiguities
No plan price, seat rate, per plan credit allotment, or credit bundle rate appears on any first party page as of 28 Aug 2026; all of it sits behind login or a sales conversation. Third party pricing roundups still circulate the withdrawn $60, $150, and $50 figures, so any number sourced off site is stale unless re-confirmed inside the product.
Cancellation / refund
Enterprise customers are billed per their contract through a dedicated account manager rather than the self serve flow. Annual billing previously carried a 10% discount on the base subscription; that figure is no longer published and is not currently verifiable.
Support SLA / resale
Enterprise adds unlimited product usage, SSO, private cloud hosting, custom credit allotments, and a dedicated account manager.
Missing data
All prices. Plan fees, the editor seat rate, per-plan credit allotments, credit bundle and auto top-up rates, and the extra phone number and concurrent call add-on rates are none of them published on a first-party page.
Related vendors
- Decagon — AI agent platform purpose-built for customer support, with deep…
- 5.Y — 5
- Ada — Agentic customer experience platform whose AI agents autonomously…
- Aisera — Enterprise agentic AI platform that orchestrates specialized agents…
- ASAPP — Generative AI for enterprise contact centers that plugs into…
- Assembled — AI customer support orchestration platform that unifies autonomous…
Alternatives to Voiceflow
The closest documented capability profiles to Voiceflow among customer support agents tracked by Agentic Index, ordered by similarity on the same 14 point evidence the rankings use. No vendor pays for placement.
- Plivo9.5 / 14Adds documented Human Oversight & Guardrails and Memory & State Persistence
- Chatbase10.0 / 14Adds documented Human Oversight & Guardrails and Memory & State Persistence
- Dialpad10.0 / 14Adds documented Human Oversight & Guardrails and Memory & State Persistence
- Kommunicate10.0 / 14Adds documented Human Oversight & Guardrails and Memory & State Persistence
- LiveChatAI9.0 / 14Adds documented Human Oversight & Guardrails and Memory & State Persistence
- BlueTweak9.5 / 14Adds documented Human Oversight & Guardrails and Memory & State Persistence
Similarity is computed from each vendor's Agentic Index coverage score evidence, axis by axis, not from the totals. How this evidence is graded