Bolna

Bolna markets an India-oriented platform for inbound and outbound multilingual voice agents, with no-code and API control, bulk calling, real-time custom API triggers, human transfer, workflow integrations, 20+ model providers and India/US or on-premises deployment. It says 1,000+ companies use Bolna across 10+ vernacular Indian languages and exposes five named customer-story cards whose metrics sometimes conflict with the dedicated stories.

The service agreement identifies WHISMURWAVE INC doing business as Bolna. Customers are responsible for communications placed through their accounts and for applicable legal compliance; the platform says it facilitates communications without monitoring or endorsing their content and prohibits fraudulent, abusive and unlawful use.

Whismurwave Inc says Bolna collects call transcripts, recordings, phone numbers and service data; account administrators can access data associated with their organization. The policy describes encryption, access controls and secure servers, says personal data is retained while the account is active or as needed, supports deletion requests by email and says inactive-account data is removed after three years.

Y Combinator lists Bolna as an active Fall 2025 company in Bengaluru with founders Prateek Sachan and Maitreya Wagh and a team size of 21. Its launch copy describes sub-500-millisecond voice agents across 10+ languages and 50+ accents, serving Indian enterprise sales, support, collections and recruitment, with 500,000+ minutes per month across BFSI, ecommerce and recruitment.

Bolna's official LinkedIn profile says it was founded in 2024, lists a self-selected 11-50 employee band, gives San Francisco as headquarters and Bengaluru as a location. A recent company post says the team grew past 40, which is not synchronized with Y Combinator's team-size figure of 21.

Co-founder Prateek Sachan says Bolna raised a USD 6.3 million seed round led by General Catalyst, with Blume Ventures, Pioneer Fund, Transpose Platform, Orange Collective, Eight Capital and angels participating; he also names Y Combinator and Upekkha as early supporters. The post says Bolna had reached 1,000+ customers and 8.8 million people in India.

Bolna's official company update confirms a USD 6.3 million fundraise led by General Catalyst and says its team grew past 40 people.

A General Catalyst portfolio job listing describes Bolna as Bangalore-based, funded with a USD 6.3 million round led by General Catalyst, and operating across 10+ Indian languages. It reports 250,000+ daily calls, 1,500+ paying customers and 15% month-over-month growth, without defining the growth denominator.

Bolna sells prepaid pay-as-you-go credits from USD 10 to USD 5,000 without commitment and displays volume pricing beginning at USD 15 for 250 minutes, equivalent to USD 0.06 or INR 5.52 per minute. A one-time pilot is 10,000 paid minutes plus a 20% bonus for 12,000 total, billed in 30-second pulses with up to 100 concurrent calls and one preconfigured agent. Enterprise pricing is custom; new accounts receive USD 5 credit and BYOK is supported.

Call cost is composed from speech-to-text duration, LLM tokens, text-to-speech characters, telephony provider and destination-country charges rounded to provider minutes, plus a flat Bolna platform fee per minute. BYOK removes Bolna-billed provider components for supplied keys but not the platform fee.

Bolna publishes outcome-oriented recruitment pricing of USD 1 per successful CV screening or role-play screening, with 30 free test minutes, and a relationship-manager add-on priced at USD 0.05 per minute.

The FAQ describes a USD 0.02-per-minute platform fee plus provider charges, up to 10 default concurrent calls on paid accounts, API-first management of agents, calls, executions, batches, phone numbers and subaccounts, team sharing and enterprise on-premises deployment. It says SIP is not yet supported, which conflicts with the newer integrations page and current SIP tutorials that explicitly support generic and named-provider SIP.

Bolna documents inbound, outbound and batch calls, human transfer, Cal.com availability and booking, custom API tools, recordings, transcripts, summaries, intent, sentiment and structured post-call data. Agent configuration is divided across Agent, LLM, Audio, Engine, Call, Tools, Analytics and Inbound controls.

Users can create or import an agent, configure prompts, knowledge, language, telephony, tools, analytics and caller matching, then test in chat, browser voice or by phone before deployment. Drafts do not apply until saved; the interface exposes estimated cost per minute and India-specific routing.

Agents can invoke customer-configured HTTP APIs during a live call, use built-in Cal.com availability and booking, transfer calls and run optional pre-call webhooks. API schemas can be entered manually or imported from cURL; the customer selects endpoints, parameters and permitted tool behavior.

A real-time execution webhook can send call events to an external system. Post-call controls generate summaries and configured structured extractions such as handoff status, lead qualification or sentiment, can be tested before deployment, and can push outputs into CRM or other systems.

Bolna provides queued, in-progress and completed call status through polling and webhooks, with raw execution payloads usable for CRM and dashboard automation and a documented fixed outbound IP.

Call history exposes execution count, total cost and duration, status, average cost and duration, recordings, transcripts, timestamped trace logs, raw JSON and CSV export; queued calls can be stopped. Filters support agent, status and date analysis.

Call controls cover noise cancellation, voicemail, DTMF, failed-call rescheduling, calling windows, final messages, silence and maximum-duration timeouts and ambient sound. The customer configures timing and retry behavior.

The batch API accepts a CSV of phone numbers and per-call context, supports multiple originating numbers and configurable retry count and intervals, and returns a batch identifier for later tracking.

A current tutorial automates outbound calls from Google Sheets, writes statuses and transcripts back through webhooks, supports batch processing, structured extraction and automatic retry.

The v2 API supports programmatic creation, listing, updating and deletion of agents, including task prompts, configuration and webhook fields.

Agents can use retrieval-augmented knowledge bases created from PDFs and URLs, attach multiple knowledge bases and operate across multilingual content; the documentation says knowledge bases support 100+ languages. This is persistent reference knowledge, not evidence of automatic episodic memory across calls.

Inbound agents can match a caller against an internal API, CSV or public Google Sheet and inject supplied name, address, preferences and prior-history fields into the prompt. Persistent customer context depends on merchant-provided systems rather than an evidenced native cross-call memory store.

This documentation surface says the platform supports more than 50 global languages and language combinations, while other current official surfaces say 10+, 40+ or distinguish 100+ knowledge-base languages. The differing figures are retained as scope-dependent rather than combined.

The audio configuration lists English, Hindi, Bengali, Assamese, Gujarati, Kannada, Malayalam, Marathi, Odia, Punjabi, Tamil, Telugu, Urdu and several non-Indian languages, allows separate speech-to-text and text-to-speech providers by language, and supports language switching during a call.

Agent setup supports per-language prompts, call-time variables, reusable prompt modules for collection, flow, sector and universal instructions, language switching, guardrails, context-based hangup and per-language human-handoff messages.

Graph Agents are explicitly beta. They model multi-step conversations as nodes and transitions, including static, router and LLM nodes, forced function calls, per-node retrieval, expression- and event-based transitions, external REST event injection and debugging.

Bolna documents inbound and outbound calling through Plivo, Exotel and Vobiz for India and Twilio across several international markets, with bring-your-own-provider credentials and SIP support subject to the provider. This current evidence supersedes but does not erase the older FAQ's no-SIP statement.

The current directory lists model integrations including Anthropic, Azure, DeepSeek, OpenAI, OpenRouter, Perplexity and Sarvam; speech providers including Deepgram, Cartesia, ElevenLabs, Rime and Smallest; telephony and SIP through Exotel, Plivo, Twilio, Vobiz, Acefone, Airtel and generic SIP; and workflow tools including Cal.com, Make, Zapier and AiSensy WhatsApp. Shopify order lookup is described as possible through custom APIs, not as a named native Shopify connector.

Bolna documents TLS 1.2+ and encrypted audio in transit, encrypted provider credentials, hashed API keys, subaccount isolation and spend/agent/number controls. Default storage is AWS us-east-1; enterprise India residency is available. Webhooks originate from a fixed IP but currently have no HMAC signature. The page instructs customers to contact Bolna about HIPAA, SOC 2 and GDPR applicability rather than asserting certification, and describes responsible-calling guardrails such as time windows, DNC and maximum duration.

Default deployment stores data in AWS us-east-1. Enterprise customers can select AWS ap-south-1 so recordings and transcripts stay in India and model inference is routed through India-hosted endpoints where providers make them available.

Enterprise customers can deploy a containerized Bolna stack with Docker and Kubernetes in their cloud or data center, keeping audio, requests, logs and transcripts in their environment. Performance and usage metrics are still sent to Bolna for billing and optimization, and APIs remain compatible.

Enterprise subaccounts provide logical isolation, independent API keys, logs, analytics and call history, with consolidated billing, usage quotas and agent/number restrictions. Dashboard lifecycle management is beta; phone and provider resources remain shared at organization level.

Organizations have Admin and Member roles. Admins have full organization management, while Members can place calls and manage their own API keys but face restrictions on destructive or organization-level changes; APIs enforce role scope.

Enterprise administrators can set organization and subaccount concurrency minimums and maximums. Excess outbound calls queue; inbound calls are never queued, so capacity behavior differs by direction.

Bolna exposes importable, production-ready voice-agent templates for ecommerce customer support, cart abandonment, COD confirmation and surveys. The COD template confirms intent and delivery details before dispatch to reduce fake or abandoned shipments; this establishes COD confirmation, not NDR or RTO execution.

The ecommerce offer covers order-status conversations, COD confirmation, recovery campaigns and support triage. It does not document native order management, carrier booking, payment processing or NDR/RTO resolution.

The importable agent is triggered from merchant cart events, calls the shopper, handles objections, can query delivery, payment and coupon data through APIs, sends an SMS or WhatsApp checkout link only after consent, records recovery tags and can transfer to a human. Discounts are limited to merchant-defined rules rather than autonomously invented offers.

The support template verifies callers, retrieves order and shipment status, executes sequential custom API calls and transfers callers with transcript and summary. With merchant-supplied commerce and returns APIs it can initiate a return label, refund or account credit; Bolna does not itself supply those systems, and no named case proves production return/refund execution.

Bolna says Awign automated technical screening and scheduling, reporting 65% lower interview cost, 100+ simultaneous interviews, roughly 50% of candidates called within three hours, roughly 90% completion after rescheduling and twice the completion from WhatsApp/SMS nudges versus cold outreach. The dedicated metric panel says 5.0 minutes longest and 0.0 minutes average, while the homepage says 17.1 and 6.8 minutes; the impossible-looking average and cross-page conflict are preserved.

Bolna says Hyreo deployed a 24x7 recruiting helpline and proactive outreach, reporting full call coverage, more than 90% complete candidate data, 60% lower outreach cost, up to 80% recruiter time saved and 12% of candidates identified at risk and engaged. The dedicated page reports 50 talk hours, 50% monthly call-minute growth, 100 hourly peak calls and 5,000 conversations; the homepage instead reports 96.55% growth and 10,000+ conversations.

Bolna says GoKwik used voice agents for cart recovery, surveys, collections and announcements, answered COD, discount and delivery questions, switched language during calls and sent WhatsApp checkout links. The dedicated page reports 250,000 ecommerce call minutes, 400,000 unique customers and 150 concurrent calls; the homepage reports 400,000+ engagements and 250+ peak concurrency. A panel value of one purpose-focused agent appears malformed and is not treated as outcome proof.

Bolna says Hypothesis ran 95,000 cart-recovery calls and recovered more than INR 2.5 crore, with agent examples using merchant-defined discounts and checkout links. The homepage reports a 24-hour turnaround and 300+ SKUs, while the dedicated panel says 250 SKUs and shows malformed revenue and turnaround values; the malformed values are not normalized.

Bolna says Futwork deployed a national support campaign for an unnamed digital-payments client, reporting 9,000 calls daily, 200 concurrent calls, 80% lower human effort and 18% of calls automatically escalated with context. Homepage cards instead report 10,000+ daily calls and 250+ concurrent calls. Example flows cover education, ticket creation and device-replacement requests, but the end client and exact systems remain undisclosed.