Back to vendors
S

Sarvam AI

Also known as: Sarvam, Samvaad, Sarvam Voice Agents, Sarvam Work Agents, Indus Cowork, Axonwise Private Limited, sarvam.ai

Visit site
Entry pricePay as you go at ₹0 a month; voice agents from ₹3.50 a minute, models includedFull pricing detail

Sarvam AI builds AI models and agents for Indian languages. Its Voice Agents platform runs phone agents in Hindi, Tamil, Telugu and other Indian languages that call APIs mid conversation to book appointments, qualify leads, offer loans and follow up on collections, with a conversational builder and AI assisted testing. Work Agents connect to more than 100 apps for scheduled work, and Sarvam deploys in its cloud, a private cloud, on premises or air gapped, with data kept in India.

Sarvam AI is a Bengaluru company, operating as Axonwise Private Limited, that builds AI models and agents for Indian languages and sells them to enterprises and government. Its own models include the Sarvam 105B language model, Saaras speech recognition for 22 Indian languages and English, Bulbul text to speech with 83 voices in its V4 Flash version, Mayura and Sarvam Translate for translation, and Sarvam Vision for documents. Customers include Axis Bank, CRED, IDFC FIRST Bank, IndiaMART, Infosys, LIC, Mahindra Finance, NABARD, SBI Life, Tata Capital and Urban Company.

Voice Agents is the main agent product, a platform to build, test and run phone agents in Hindi, Tamil, Telugu, Bengali, Marathi, Gujarati, Kannada, Malayalam, Punjabi and other Indian languages. Teams describe the agent to Genie, a conversational builder, link a knowledge base and launch on a rented or existing number in minutes.

Agents handle inbound and outbound calls and event driven campaigns for appointment booking, cart recovery, lead qualification, preapproved loan offers, order status and collections, calling APIs mid conversation to fetch information, update records and complete tasks. Branching flows move between specialist agents, tools and follow up actions under shared goals, with waits, switches and voicemail detection, and responses come back in under 500 milliseconds. Tata Capital runs consumer loan calling on Sarvam in English and 10 Indian languages at more than 2 million interactions a month, passing judgment heavy conversations to human agents.

Work Agents takes on recurring work and projects. A team member describes the job, defines the checks the output must pass, adds documents, connectors and rules, and shares the agent with the team. Agents connect to Notion, Slack, HubSpot, Google Drive, Linear, GitHub, Jira and more than 100 other apps, run on daily, weekly or monthly schedules, split jobs among sub agents and return output for review, with their data kept in India.

Indus Cowork, an AI teammate that waits for approval before anything ships, is not generally available yet. Developers can also build on Sarvam's speech, translation, document and language model APIs, with Python and JavaScript SDKs, an OpenAI compatible endpoint that also serves open models such as GLM, Gemma and DeepSeek, and guides for LiveKit and Pipecat.

Sarvam runs as a managed cloud, a private cloud, on premises, hybrid or air gapped, with production data held in India. It holds ISO 27001 certification and a SOC 2 Type II report, offers customer managed keys and configurable retention, and never uses one customer's data to train models for another.

Voice agents start at ₹3.50 a minute on pay as you go, models included, falling to ₹3.00 on Business, with plans from ₹9,999 a month and custom enterprise pricing.

Sarvam suits Indian banks, lenders, insurers and public services that need voice agents in regional languages with data kept in the country, while teams focused on markets outside India will find a product built around Indian languages and residency.

Vendor details

Canonical URL

https://www.sarvam.ai/

Category

Voice agent

Subcategory

Indian language voice agents and work agents on Sarvam's own models

Company status

independent

Use cases & customers

Primary use cases

Run multilingual phone agents for collections, loan offers and lead qualificationBook appointments and answer order status calls in Indian languagesAutomate recurring reports and research with scheduled Work Agents

Target customers

Indian banks, lenders and insurersConsumer and commerce companies serving regional language customersGovernment and public service programs

Deployment options

Sarvam managed cloudPrivate cloud (VPC)On premisesHybridAir gapped

Integrations

Voice agents call APIs during calls and connect to telephony, CRM platforms and databases, Work Agents connect to Notion, Slack, HubSpot, Google Drive, Linear, GitHub, Jira and more than 100 other apps, and Sarvam's model APIs work with LiveKit, Pipecat, Twilio, Exotel, Vapi, n8n and LangChain.

In practice

A lender's voice agent calls borrowers with overdue payments in Tamil and records each repayment promise in its systems.

An online marketplace runs outbound calls in Hindi to recover abandoned carts and answer order status questions.

An operations team schedules a Work Agent to pull spend and renewal dates every month and assemble a vendor review.

Agentic Index coverage score

11.5 / 14 capabilities · 82%

Integrations & Tool Calling Full

Voice agents call APIs during live calls to fetch information, update records and complete tasks, and connect to telephony systems, CRM platforms, databases and other business tools. Work Agents read data and act through connectors to Notion, Slack, HubSpot, Google Drive, Linear, GitHub, Jira and more than 100 other apps, sending messages, accepting calendar invites and pushing code updates.

Sourcesarvam.ai/products/work-agentsread 2026-11-20

Workflow Orchestration Full

Branching call flows move between agents, tools and follow up actions based on the caller's intent and the outcome of each step, with waits, switches and handoffs from one agent to another. Goals set outcomes in plain language that several specialist agents share, and in Work Agents a lead agent splits a job among sub agents with their own tools and instructions, then combines the results.

Sourcesarvam.ai/products/voice-agentsread 2026-11-20

Knowledge Grounding & RAG Full

Voice agent builders link a knowledge base and write instructions for each agent, so calls draw on the team's own material. Work Agents use the documents a team adds as context, from one document to one lakh of them, and skills can require source citations on every task, so answers on recurring work, reports and research point back to the company's own documents.

Sourcesarvam.ai/products/work-agentsread 2026-11-20

Human Oversight & Guardrails Full

Every Work Agents run returns output for the team to review, skills can require review before anything is published, and admins choose which apps and documents each agent can reach and when it can create sub agents. On calls, Tata Capital moves nuanced, judgment heavy conversations from Sarvam's voice agents to human agents within a human in the loop framework.

Sourcesarvam.ai/products/work-agentsread 2026-11-20

Security, Identity & Governance Full

Sarvam holds ISO 27001:2022 certification and a SOC 2 Type II report covering security, availability and confidentiality. Data is encrypted with AES 256 at rest and TLS 1.2 or higher in transit, customers can bring or manage their own keys, retention is configurable per data category with an ephemeral mode, and one customer's data never trains models for another.

Sourcesarvam.ai/trust-centerread 2026-11-20

Observability & Auditability Full

Voice agents come with call analytics, transcripts, call logs, conversation patterns, performance metrics and operational monitoring in production, and the analytics export to a team's existing visualization tools. Model, prompt and evaluation versions are tracked with an audit trail, and every access to data, including personal information, is logged for audit.

Sourcesarvam.ai/products/voice-agentsread 2026-11-20

Memory & State Persistence Partial

Voice agents use customer context on each call, configured by the team, and keep that context and memory private, with control over how an agent uses them. A retention period or a store of what an agent learns about each caller across calls does not appear, and Work Agents rely on the documents, connectors and rules a team adds rather than remembered runs.

Sourcesarvam.ai/products/voice-agentsread 2026-11-20

Deployment & Data Residency Full

Sarvam deploys as its managed cloud, a private cloud in the customer's VPC, on premises, hybrid or air gapped, with single tenant setups offered. Production runs in Microsoft Azure's Central India region with language model inference on Yotta and NxtGen infrastructure, keeping data in India, and Saaras, Bulbul and Sarvam Vision can run as SageMaker endpoints in a customer's own AWS account.

Sourcesarvam.ai/trust-centerread 2026-11-20

Prebuilt Agents / Templates / Packs Partial

Voice agents come with worked use cases for appointment booking, cart abandonment recovery, lead qualification, preapproved loan offers, order status and collections, with demos for merchant support, citizen services and sales objections. Work Agents start from a ready made agent or a described job, but no library of named, packaged agents appears for teams to adopt as is.

Sourcesarvam.ai/products/voice-agentsread 2026-11-20

Triggers & Channel Coverage Full

Voice agents take inbound and outbound calls and run event driven campaigns on numbers rented through Sarvam with KYC or brought from a team's own telephony provider. Work Agents run on daily, weekly or monthly schedules without anyone starting each run, and people can also start them by asking a question in a Slack channel where an agent is deployed.

Sourcesarvam.ai/products/voice-agentsread 2026-11-20

Model Flexibility & Routing Partial

Sarvam's own speech and language models sit behind its agents, with voice agent minutes priced to include them, and enterprise deployments can bring their own model or use Sarvam's. Choosing a model per agent or routing between models does not appear among the features of Voice Agents or Work Agents, while the separate Token Factory API serves GLM, Gemma and DeepSeek beside Sarvam 105B.

Sourcesarvam.airead 2026-11-20

APIs / SDKs / MCP Extensibility Full

REST APIs cover speech to text, text to speech, translation, document intelligence and chat completions with tool calling, with Python and JavaScript SDKs and an OpenAI compatible endpoint. An official MCP server exposes those APIs as tools, including a voice agent loop, and guides cover building voice agents on LiveKit and Pipecat with Twilio, Exotel and Vapi. Hosted Voice Agents are built through Genie.

Sourcedocs.sarvam.ai/api/developer-tools/mcpread 2026-11-20

Testing, Debugging & Optimization Full

Genie runs AI assisted test cases and stress tests conversation flows before a voice agent goes live, and production calls are scored with evaluations alongside the call analytics. In Work Agents, the builder defines what success looks like, including the checks each output must pass, before sharing the agent with the rest of the team to run the same process.

Sourcesarvam.ai/products/voice-agentsread 2026-11-20

Browser / Computer-use Not documented

Sarvam's agents work through phone calls, APIs and app connectors, and Work Agents research topics and assemble reports and slide decks from connected sources. Operating a web browser or a desktop on a person's behalf does not appear among the capabilities of Voice Agents, Work Agents or the gated Indus Cowork teammate, which acts through authorized apps.

Sourcesarvam.ai/products/work-agentsread 2026-11-20

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

Pricing

Pay as you go at ₹0 a month; voice agents from ₹3.50 a minute, models included

Voice agent minutes and per unit API usage, with monthly plans that lower the rates

Trial available

Included quota

₹100 of free credits on sign up.

What is public

Voice agents cost ₹3.50 a minute on pay as you go and ₹3.00 on Business, end to end with the models included. Plans run Pay as you go at ₹0, Pro at ₹9,999, Growth at ₹49,999 and Business at ₹1,99,999 a month, with Enterprise priced on request.

Billing mechanics

Every API is metered per unit, such as speech to text at ₹0.49 a minute and Bulbul V3 text to speech at ₹3.00 per 1,000 characters on pay as you go, and higher plans lower each rate. All prices are in Indian rupees.

Sales call required

Mixed (some tiers require a call)

Free / trial

₹100 of free credits on sign up

Lowest paid plan

Pro, ₹9,999 a month

Commercial notes

Enterprise adds custom concurrency and throughput, higher rate limits, volume discounts on credits, dedicated support and SLAs. Work Agents are quoted on team size and usage.

Missing data

Sarvam sets Enterprise rates and Work Agents prices in each agreement, and Indus Cowork is sold through sales while it is not generally available.

Agentic Index verified 2026-11-20

Contact us

Found a vendor we missed? Have feedback on the index? We'd love to hear from you.