Ultravox
Multimodal LLM that takes speech input directly, without a separate speech recognition stage, for real-time voice agents.
Infrastructure by Ultravox (Fixie)
IndexedMIT★ 4,583
Get started
Setup steps are in the repository. Website · Docs · Repository
About
Ultravox extends open-weight LLMs with a multimodal projector that converts audio directly into the input space of the LLM, so no separate speech recognition stage is needed. The default model is built on Llama 3.3 70B, with an 8B variant on Hugging Face, and its weights can be downloaded there. The team also offers managed real-time APIs for running voice agents.
Use cases: voice-agents, model-inference
Reviews
Write a reviewNo reviews yet. Installed it? Reviews come from verified installers only. How reviews work
Fact sheet
- Category
- Voice · speech to speech api
- Protocols
- sipultravox.ai
Plans
| Pay as You Go | $0 | 30 free minutes, then $0.05/min; up to 5 concurrent calls; SIP at $0.005/min; no surge pricing |
|---|---|---|
| Pro | $100/month | annual-rate price; no hard concurrency caps, outbound call scheduler, 5 custom voices, 20 RAG corpora; $0.05/min |
| Enterprise | Contact sales | custom; priority SLA, org support, customizable |
From the vendor's pricing page, read Oct 24, 2026.
Facts read from the vendor's own pages by claude-haiku-5-5 on Oct 24, 2026; a fact is shown only with the page it came from.
Signals
- Stars
- 4,583
- Forks
- 385
- Open issues
- 54
- Last commit
- Dec 12, 2025 (10 months ago)
- Last release
- Aug 18, 2025 (1 year ago)
From GitHub (fetched Oct 24, 2026).
Details
- License
- MIT
- Repository
- github.com/fixie-ai/ultravox
- Website
- www.ultravox.ai
- Docs
- docs.ultravox.ai
- First seen
- Oct 24, 2026
- Last checked
- Oct 24, 2026
Cite this page
Ultravox. Agent Mart. https://agentmart.dev/infrastructure/ultravox (last updated 2026-10-24)