fusion-runtime
github.com/samarthurs18/fusion-runtime- Category
- AI Agents
- Rank
- No. 1894Tools index
Previous survey · No. 1818 ·
- Pricing
- Open Source
- Type
- TOOL
- Use case
- Agent Building
- GitHub
- 4 stars
- Latest release
- v0.1.4
- Date
About
fusion-runtime is a self-hosted voice agent runtime that streams speech-to-text, an LLM, and text-to-speech into each other within a single process, so a reply starts playing before generation finishes. On an RTX 3090 running a 7B model it reports about 490ms of processing after a turn ends (991ms from the user's last syllable) at roughly 127 tokens/sec, and it supports mid-sentence interruption.
At a glance
fusion-runtime runs speech recognition, a language model and speech synthesis together in a self-hosted voice pipeline. It is worth evaluating if you want to measure and tune the stages of a voice agent on infrastructure you operate.
Good fit
Developers prototyping a spoken interaction with control over runtime and model choices.
Consider something else when
A hosted service is simpler when you cannot operate the models or do not need control of each stage.
Start here
Use public or synthetic audio for a short exchange. Measure when transcription arrives and when the first reply audio starts, then inspect what happens on a pause.
Cost and requirements
The project documents Python 3.11 through 3.13. Model weights and compute needs depend on the selected components.
Limits to know
End-to-end latency depends on the slowest stages and hardware. Validate each component’s data path and behavior on representative audio.
Sources checked 2026-11-05. Source-based assessment.
Why it made the leaderboard
Gives builders a self-hosted alternative to hosted voice-agent APIs with real performance data and mid-sentence interruption handling for latency-sensitive local deployments.
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.