Vapi launches Vapi for CX, a voice agent toolkit for customer experience teams

The problem Vapi says it is addressing

Vapi opens its announcement with a figure from ACSI. By ACSI's estimate, companies spend well over $100 billion a year on customer experience with no detectable return. The national score, according to the post, sits exactly where it was in 2013. Complaints reached record levels in the first quarter of this year, up 16%, while corporate profits rose 15% over the same period.

Vapi's explanation is that each wave of contact center technology, from automatic call distribution in the 1970s to IVR, offshoring, web self-service and chatbots, was measured on the same numbers: cost per contact, average handle time and containment rate. Those numbers improved, the company says, but satisfaction stayed flat because nobody optimized for whether the customer's problem got solved.

Vapi describes voice AI as the first wave that could change this. Large language models let an agent hold a real conversation, understand an unstructured request and take an action in another system. According to Vapi, many deployments still point that capability at the old metric and simply measure how many calls never reach a human.

How much of the agent teams should own

The post compares three routes to a voice agent. Renting a managed agent works on day one, Vapi says, but the model is chosen for you, the orchestration is closed and the behaviors you most want to change are decided by the vendor. A custom managed service raises that ceiling without removing it, since changes are only possible inside the provider's proprietary sandbox and every adjustment becomes a ticket and a wait. Building entirely in house gives engineers full control, but also leaves them owning telephony, turn-taking, interruption handling, fallback plans, concurrency, scalability and compliance.

Vapi's pitch is that ease should be a layer rather than a replacement. Its example is Vapi Composer, a natural language way to build voice agents. A CX team can build a working agent through it without touching a model parameter, yet the result is the same config object that engineers reach through the API, the CLI and an MCP server. Hundreds of settings on a single assistant remain editable. Vapi says it manages the infrastructure layers, while customers tune the parts that are their business: prompts, logic, tools, integrations, data and model choice.

Build and deploy

For the build stage, Vapi says a CX team can describe an agent in Composer, which writes the prompt, provisions the number and wires up the tools. A model preset bundles a transcriber, model and voice into one choice. Engineers can instead work through the Vapi MCP server and the published OpenAPI spec, so a coding agent writes against real API definitions. Both routes produce one versionable config object.

The toolkit also includes Squads, which Vapi says reduce hallucinations, prompt bloat and latency by routing callers between specialized assistants mid-call with silent handoffs. Tool calling can be made deterministic with argument schemas, static variables and tool rejection plans.

Vapi cites Amazon Ring as a customer. According to the company, Ring went from zero to production in two weeks and now runs 100% of its inbound volume on Vapi, after evaluating more than 40 vendors. Vapi says Ring picked it in part because its CX and Ops teams can tune the agent without depending on engineering.

For deployment, a SIP trunk lets Vapi sit in front of existing systems, so ACD, routing, reporting and workforce management stay in place. Calls the agent cannot resolve warm transfer to a human with full context. Outbound use cases include reminders, delivery notifications, callback deflection and renewals. Vapi lists four safeguards:

• Evals: mock conversations in which each assistant turn is judged for the correct outcome.
• Versioning: autosave to drafts, with versioning and rollback.
• Concurrency and fallback: peak-hour capacity, plus fallback plans for model, voice and transcriber.
• Compliance: SOC 2, HIPAA, PCI, zero data retention, EU residency and SSO.

Observing performance

For monitoring, Vapi describes Scorecard, which grades 100% of calls against criteria the customer defines. Examples given include whether the agent verified identity before touching the account, whether it resolved the issue without transferring, and whether it offered escalation when policy required it. The post also mentions structured outputs that pull out the fields a team cares about, though the available text cuts off before the rest of the observation features are described.

What it means for CX teams

The launch is aimed at organizations that want contact center staff closest to the customer to change agent behavior without waiting on engineers, while keeping engineering-level access to the underlying settings. Vapi's stated argument is that control over tools, schemas and behavior sets how much value a single interaction can produce, and that resolving a problem end to end requires reading entitlement logic, writing to a company's systems and confirming the change. The source does not give pricing or availability details. Vapi also notes that VapiCon takes place on November 11-12.