3/ Developer Updates •– voice agent tests are now live on the platform. Users can describe and build tests there docs.sarvam.ai/conversations… •– sarvam launched an MCP server to connect voice agent building workflows to supported harnesses ↳ docs.sarvam.ai/conversations… •– voice agent platform's build, deploy and monitor flow includes a dashboards API, call transfers, inbound webhooks and phone number rotation ↳ sarvam.ai/products/voice-age… •– team announced Sarvam vision 2.1, highlighting its OCR, math and tiny text benchmark performance ↳ sarvam.ai/blogs/sarvam-visio… •– new discord channels and build challenges are planned for the community but no community links was shared

Q: Can Sarvam help take a project further? A: Sarvam may help with credits and mentorship. team also mentioned its startup program Q: Does 105B support cached prompts? A: not currently team offered to check and follow up Q: How is real time speech priced? A: team said pricing should match Saaras v3/v4 and depend on audio duration. they were investigating a dashboard discrepancy and said they would confirm whether silent open websocket time is billed Q: Where can users send feedback? A: through the community corner Q: Is dark mode available in Indus? A: not yet, android and iOS apps have dark mode Q: How was Sadak built? A: first prototype took about 6 - 8 hours with more updates before the finals. The team used claude opus for three.js code and later updated the game world using Indian imagery and map data Q: Was moving from v3 to v4 difficult? A: No the team said the switch was simple and key-term prompting improved recognition of important words

1/ Sarvam Campus update campus tour covered 3 universities: IIT Madras, NIT Trichy and SRM about 200 projects were built in 45minutes build challenges the plan is to visit 35 more campuses and build a community across the country

Joined @SarvamAI dev office hours with @VinayakGavariya today with lots of exciting updates •- real-time STT api updates •- saras v4 key term prompting •- new model and api updates •- community project showcase •- developer Q&A let's recap!

3/ Q&A session Q: Any update on the upcoming trillion+ parameter model? A: no details have been shared yet. The team has asked to keep an eye on the announcement channel Q: Voice cloning api? A: voice cloning is available on the platform but api is not yet available Q: Can sarvam models be deployed privately? A: on prem deployment is available depending on the use case Q: Any update on sarvam champions? A: this is still in the works. the program will include creators and city leads while the team is currently focusing more on the sarvam campus program

Today i joined @SarvamAI Dev Office Hours for an awesome session with @VinayakGavariya - latest dev updates and new APIs & models - hackathon project demos - developer Q&A Let's recap!

3/ Q&A session Q: How often are Dev Office Hours held? A: Biweekly, every other Friday Q: Are word-level timestamps available in Saras V3? A: No, and there is no confirmed ETA yet Q: Are small open source models on the roadmap? A: not currently, the focus is on improving existing models Q: When will Sarvam Code be available? A: the beta is closed with GA expected later Q: Is key-term prompting coming to Saras? A: yes, it is planned for Saras V4 to improve domain specific transcription Q: What is Sarvam Champions? A: a program for creators and city leads with early product access, direct team interaction, community access and resources