Last updated: 2026-10-14
Bolna AI reported more than 200,000 daily calls in January 2026, up from about 1,500 a day after its first commercial deployment in May 2025. The Bengaluru platform, founded in 2024, orchestrates speech recognition, language models and speech synthesis so enterprises can run phone agents in Hindi, Tamil, Telugu and Hinglish.
About Bolna AI
Bolna AI is a voice agent platform from Bengaluru that lets a business answer and place phone calls with AI. Maitreya Wagh (CEO) and Prateek Sachan (CTO) founded it in 2024, it joined Y Combinator's Fall 2025 batch, and a $6.3 million seed round led by General Catalyst followed in January 2026, according to the company's announcement. The company names Varun Beverages, Spinny and Snabbit among its customers.
Each call runs through a pipeline of speech-to-text, a language model and text-to-speech, and Bolna's job is to pick and connect those parts. Transcription options include AssemblyAI and Deepgram, voices can come from ElevenLabs as well as Indian providers such as Sarvam, and the language model can be an OpenAI or Google model, including OpenAI models served through Azure. Telephony connects through Twilio, Plivo, Exotel, Vobiz or your own SIP trunk, so you can change one provider without rebuilding the agent.
The docs describe six ways to build an agent: Agent Studio, an agent builder that drafts a call-ready agent from a short description, graph agents, workflows, the REST API, and a CLI or MCP server. The homepage shows four starter agents: reminders for EMIs and collections, inbound customer support, lead qualification and announcements. Mid-call actions use custom API triggers, a call can be transferred to a human, and results can flow into n8n, Make or Zapier. Calling guardrails cover time-of-day windows, do-not-call lists and maximum call length.
Where the vendor's own material disagrees, test before you commit. The homepage cites 10+ languages and replies under 300 ms, while the docs cite 40+ languages and sub-600 ms responses. The security documentation lists no SOC 2, ISO 27001 or HIPAA attestation and points buyers to support, although it reports an independent VAPT assessment that the vendor says earned an A+. It also notes that webhooks carry no HMAC signature, so you verify Bolna's three source IP addresses instead. Pricing is usage based per minute and is covered in the cost section below.
For comparison, Vapi and Retell AI are the closest developer-first platforms on HokAI. Bolna is the narrower choice, strongest when the calls are in Indian languages and the data should stay in India, and its MIT-licensed repository lets technical teams inspect or self-host the orchestration layer. HokAI's guides to the best AI voice assistants, customer support tools and recruiting tools cover the wider field, and Smart Match can shortlist a platform for your use case.
Pricing
06 per minute, with speech-to-text, the LLM and text-to-speech bundled when the agent uses Bolna's preferred models; telephony and a platform fee are added per minute. 5 cents per minute. New accounts get $5 in free credits.
Pilot plan (one-time purchase, billed in 30-second pulses): $300 for 6,500 minutes or $500 for 12,000 minutes, with up to 25 concurrent calls, enterprise features and one month of dedicated support. Enterprise pricing is custom and volume based. Bringing your own provider keys (BYOK) lowers cost.
Rival platforms such as Retell AI also bill by the minute. ai/pricing on 2026-10-14.
| Tier | Monthly price | What it includes |
|---|---|---|
| Pay-as-you-go | Free | |
| Pilot (one-time purchase) | Custom | |
| Enterprise | Custom |
Key Features
- Six ways to build: Create agents in Agent Studio, with an agent builder that drafts one from a short description, as graph agents or workflows, through the REST API, or with a CLI and MCP server.
- Bulk outbound campaigns: Runs campaigns that place thousands of AI calls at once, with do-not-call lists, time-of-day windows and maximum call length set through calling guardrails.
- Choice of speech and language models: Connects 20+ speech, language and voice models, switches models per call, and accepts your own provider keys.
- Mid-call actions and human handoff: Calls external APIs in real time during a conversation and transfers the call to a human agent when needed.
- Bring your own telephony: Attaches lines from Twilio, Plivo, Exotel and Vobiz, or your own SIP trunk, so calls use your own numbers.
- Data residency and self-hosting: Offers India and USA data residency, routing through Indian servers, and a Docker or Kubernetes guide for on-premise deployment.
Pros
- Designed first for Indian languages and accents, including Hinglish code-switching, rather than added to an English-first stack.
- The core repository is MIT licensed, so technical teams can read the orchestration code or self-host it.
- Self-serve credits with no commitment mean a pilot needs no sales call, and a named-customer list backs the claims.
Cons
- Webhook payloads and tool-call requests carry no HMAC signature in the current version, so teams must whitelist the three Bolna source IPs instead.
- Homepage and docs disagree on headline specs for languages and latency, so test both on your own calls before signing.
- The security docs name no compliance certificate, and buyers needing one must ask support.
- Telephony and a platform fee sit on top of the bundled model rate, so the all-in cost per minute is above the headline price.
Data Handling
- Compliance
- None published; independent VAPT assessment reported by the vendor
Frequently Asked Questions
What does Bolna AI actually cost?
Bolna charges by the minute: the entry rate is $0.06, which covers transcription, the model and the voice while you stay on its bundled models, and telephony plus a platform fee come on top. Bigger top-ups lower that to roughly 4.5 cents, and the Pilot plan sells 6,500 minutes for $300 or 12,000 for $500, both with 25 concurrent calls. Enterprise quotes are custom.
Can you use Bolna AI without paying?
There is no permanent free plan. Every new account receives $5 in free credits to test the platform, and the docs offer a live demo agent you can talk to without signing up. When credits run out, agents stop placing calls until you top up, though automatic top-ups and low-balance notifications can be switched on.
What are Bolna AI's closest competitors?
On HokAI the nearest rivals are Vapi and Retell AI, both billed by the minute of call time. ElevenLabs is the pick if voice quality is the main requirement. Choose Bolna when Indian-language coverage, India data residency or a self-hostable MIT-licensed core is the deciding factor.
How does Bolna AI compare to Vapi in 2026?
Both bill per minute with no contract, but they price differently: Vapi lists $0.05 per minute for orchestration with speech, language-model and voice costs billed on top, while Bolna's headline rate bundles its preferred models and adds telephony and a platform fee. Vapi's HokAI record describes a bring-your-own-model layer, whereas Bolna's differentiators are Indian-language focus, India data residency and an open repository. Pick Bolna for Indian-market calling and Vapi for a global, model-agnostic build.
How do you set up Bolna AI?
Sign up on platform.bolna.ai, then create an agent from a template or a short text description in Agent Studio, and clone a demo agent if you want a starting point. Link a phone line from Twilio, Plivo, Exotel or Vobiz, or point your own SIP trunk at it, then trigger a test call from the dashboard or with a POST to the call endpoint of the REST API. Developers can drive the same steps through the API, the CLI or the MCP server described in the docs.
Top Alternatives
- Vapi: Pick Vapi if you want a global, model-agnostic voice build; pick Bolna if your callers speak Indian languages and data should stay in India.
- Retell AI: Pick Retell AI if you want its published per-minute stack and enterprise discounts; pick Bolna if bundled models and Indian-language focus matter more.
- ElevenLabs: Pick ElevenLabs if realistic voice output is the priority; pick Bolna if you need a phone-call orchestration layer with ready-made agents.