Retell AI Pricing, Free Tier, and How It Builds Voice Agents
How Retell AI's free tier, concurrency limits, and no-code builder work, based on a hands-on build of a phone-based AI voice agent.
What is Retell AI?
Retell AI is a platform for building AI voice agents that hold phone conversations, answer questions, and route calls the way a human call center agent would. It supports voice, chat, SMS, and API deployment, so the same agent logic can show up as a phone line, a web widget, or a backend integration. The pitch is a call center you can build without writing code, then scale with the monitoring and compliance tooling that enterprise buyers expect.
The platform is built around a prompt-driven agent creation flow. You describe the agent’s role, its boundaries, and its knowledge, and Retell handles the speech-to-text, text-to-speech, and turn-taking logic underneath.
TL;DR
- Retell AI lets you build a working phone or web voice agent from a single prompt template, with no code required to get a first version live.
- The builder supports a knowledge base you can populate with text, uploaded files, or web pages, and the first ten knowledge bases come at no extra charge.
- A hands-on test showed the agent handling mid-sentence interruptions gracefully and sticking to a stated boundary (refusing personalized financial advice) even when pushed.
- The analytics dashboard tracks call duration, latency, concurrency, and sentiment, with a test call logging around 39 seconds and roughly 1,200 milliseconds of latency.
- An AI quality assurance layer scores every call automatically for audio quality, hallucinations, resolution, and sentiment, instead of relying on manual spot checks.
- Default concurrency is capped at 20 simultaneous calls, with quota increases available for enterprise volume, backed by a stated 99.9% uptime target and auto-scaling.
- Compliance coverage includes HIPAA (with a BAA), SOC 2 Type 1 and Type 2, and GDPR, plus self-serve BAA and DPA signing and configurable data retention from one day to two years.
Other agents start typing. Remy starts asking.
Scoping, trade-offs, edge cases — the real work. Before a line of code.
How do you build a voice agent in Retell AI?
The creation flow starts with a “create agent” button, which leads to a choice of agent types, including a single-prompt voice agent. From there, the builder presents a large instruction box where you define:
- The agent’s role (for example, a financial independence education assistant)
- Hard rules or boundaries it must never cross (such as not giving personal financial advice)
- The topics it’s allowed to cover
- How it should talk and structure the call
Below that, you set a welcome message. You can let the AI speak first with either a fixed greeting or a dynamic one, so the agent opens every call the same consistent way instead of waiting passively for the caller to start talking.
This structure means the heavy lifting of “personality” and “compliance” happens in plain language, not in a flowchart of nodes. That’s a meaningful difference from older IVR-style builders, where boundaries had to be hardcoded into call trees.
How does the knowledge base work?
Retell AI lets you attach a knowledge base to any agent so it answers from specific, grounded information rather than general model knowledge. You can add content three ways: paste text directly, upload files, or point it at web pages. Each knowledge base gets a name, and you can attach multiple knowledge bases to a single agent.
For testing purposes, adding a block of custom text (for instance, detailed information about the 4% safe withdrawal rate and other financial independence concepts) is enough to ground the agent’s answers in specific facts instead of generic responses. The first ten knowledge bases are included at no extra charge, which matters for anyone testing multiple agents or use cases before committing to a paid tier.
Once the knowledge base is saved, you return to the agent page, click add, and select it from the list. The agent then pulls from that source when answering relevant questions.
Does Retell AI handle interruptions and boundaries well?
In a hands-on test, the agent was interrupted mid-answer while explaining the safe withdrawal rate concept, and it recovered cleanly without breaking the flow of conversation. Handling interruptions gracefully is one of the harder problems in voice AI. Many voice agents either talk over the caller, freeze, or lose context when a person cuts in, so this is a practical marker of quality rather than a cosmetic feature.
The same test pushed the agent toward giving specific investment advice (“exactly what to buy”), and it declined, restating that it can’t give personal financial advice and offering to explain general concepts like index investing instead. That’s the boundary set in the original prompt holding up under a direct challenge, which suggests the instruction-following in Retell’s single-prompt agents is reasonably reliable for guardrail-style rules.
What does the analytics and monitoring stack include?
One coffee. One working app.
You bring the idea. Remy manages the project.
Once an agent is published, Retell AI’s analytics dashboard tracks call counts, duration, latency, concurrency, and user sentiment. A single test call showed a duration of about 39 seconds and roughly 1,200 milliseconds of latency, giving a concrete sense of the response speed callers experience.
Beyond basic analytics, the platform includes:
- Live monitoring, which lets you watch calls happening in real time, effectively giving you a call center floor view when multiple agents are active.
- AI quality assurance, which scores every call automatically across audio quality, hallucination rate, resolution, and sentiment, instead of requiring manual review of a sample.
- Alerting, which notifies you by email or webhook when something goes wrong, such as payment failures, concurrency spikes, or provider errors.
- Integrations with existing contact center and CRM platforms, including Salesforce and HubSpot, with more integrations reportedly in progress.
This combination (real-time visibility, automated QA, and alerting) is what pushes Retell AI from “voice chatbot demo” toward something a business could actually run a support or sales line on.
What are Retell AI’s usage limits and uptime guarantees?
Retell AI states a 99.9% uptime target along with auto-scaling infrastructure. The default concurrency limit is 20 simultaneous calls per account, which is a reasonable ceiling for testing or a small operation but will bottleneck quickly for a business running real call volume. Enterprise customers can request quota increases beyond that default to match their expected traffic.
Anyone evaluating Retell AI for production use should treat the 20-call default as a planning number: if your expected peak call volume exceeds that, budget time to negotiate a higher quota before launch, not after you’re already fielding dropped or queued calls.
Is Retell AI compliant enough for regulated industries?
Retell AI covers several major compliance frameworks: HIPAA (available with a signed Business Associate Agreement), SOC 2 Type 1 and Type 2, and GDPR. Businesses can self-sign a BAA and a Data Processing Agreement directly, and per-agent data retention is configurable, ranging from as little as one day to as long as two years, with PII controls included.
One caveat worth flagging for international or regulated deployments: Retell AI has stated it does not run services inside the EU. For organizations with strict data residency requirements, that’s a detail to check against your own legal and regulatory obligations before committing, rather than assuming coverage. Reaching out to Retell directly for guidance on data residency is the sensible move if this applies to your use case.
Is Retell AI worth it?
For teams that need a voice agent live quickly, the combination of a no-code prompt builder, an easy-to-populate knowledge base, and genuinely useful call analytics makes Retell AI a strong starting point. The interruption handling and boundary adherence held up in testing, which are two of the most common failure points for conversational voice AI. The free tier and the ten free knowledge bases make it low-risk to prototype an agent before deciding whether to scale.
The tradeoffs to weigh are the default 20-concurrency cap, which will require an enterprise conversation for any real call volume, and the compliance fine print, particularly around EU data residency, which needs to be checked against your specific regulatory requirements rather than assumed.
Frequently Asked Questions
Built like a system. Not vibe-coded.
Remy manages the project — every layer architected, not stitched together at the last second.
Does Retell AI have a free tier?
Yes. Retell AI offers a free tier to start building and testing agents, and the first ten knowledge bases are included at no extra charge, which covers most early prototyping needs.
What is Retell AI’s default concurrency limit?
The default is 20 concurrent calls per account. Businesses expecting higher call volume need to request a quota increase for enterprise-level usage.
Can Retell AI agents handle being interrupted mid-sentence?
In hands-on testing, yes. The agent handled a mid-answer interruption smoothly and continued the conversation naturally, which is a common weak point for voice AI platforms.
Is Retell AI HIPAA and GDPR compliant?
Retell AI supports HIPAA compliance with a signed BAA, along with SOC 2 Type 1 and Type 2 certification and GDPR coverage. It also allows self-signing of BAAs and DPAs and configurable data retention periods.
Does Retell AI operate in the EU?
Retell AI has stated it does not run services inside the EU, so organizations with strict data residency requirements should confirm this against their own compliance obligations before deploying.