# uRun

> uRun is a real-time decision model for voice agents. It listens to live conversations and answers questions up to 12.5 times per second, so developers can control exactly when their agents speak.

- Type: AI tool
- Page: https://completeaitraining.com/ai-tools/urun/
- Published: 2026-11-18
- Updated: 2026-11-19
- Author: Jeroen Erne
- Jobs: IT and development, Customer support
- Topics: API, Developer Tools
- Reading time: 2 min

### About uRun

uRun is a platform that hosts real-time decision models for voice and video agents. Its first model, Sense, processes streaming audio and video to answer developer-defined questions up to 12.5 times per second. The model makes decisions in under 80ms without first converting speech to text, which lets agents handle turn-taking, interruptions, and other timing-sensitive behaviors directly from the media stream.

### Review

Voice agents often struggle with conversational timing-jumping in during natural pauses or freezing when a user interjects. Sense approaches this by continuously evaluating the live stream against criteria the developer sets, returning calibrated probabilities that code can act on. The tool is available now with SDKs for Python and TypeScript, and new accounts receive 1,000 free credits to test it.

#### Key Features

- Streaming decision model that evaluates audio, video, and text inputs continuously rather than in single shots
- Sub-80ms response time, enabling real-time orchestration of turn-taking without speech-to-text preprocessing
- Custom question definitions-developers write their own criteria and subscribe to a stream of probability scores
- Multilingual turn-taking support, with performance varying by language and tunable thresholds for different cadences
- Python and TypeScript SDKs for real-time sessions, plus single-request mode for exploration

#### Pricing and Value

uRun offers a free tier that includes 1,000 credits for new accounts. The reference content does not specify pricing beyond these initial credits, so ongoing costs after the free credits are exhausted are not yet defined publicly. Developers can evaluate the model's behavior within the free allocation before any paid commitment is required.

#### Pros

- Decisions operate directly on raw audio and video, skipping the latency of transcription-then-LLM pipelines
- Continuous stream awareness means the model tracks what happened before while evaluating the current 80ms window
- Developer-defined questions allow precise control over what the model watches for in a conversation
- SDKs in two languages and a single-request endpoint lower the barrier to initial testing
- Threshold tuning per language helps adapt turn-taking to different speech rhythms

#### Cons

- Turn-taking accuracy varies across languages, with some languages described as much weaker than others
- Pricing structure beyond the 1,000 free credits remains undefined, making cost projection difficult
- The tool is not well suited for teams that need a fully managed turn-taking solution with pre-built conversational policies-Sense requires developers to define their own questions and thresholds

Sense fits teams that already have voice agent infrastructure but keep hitting walls with silence timers and rule-based turn detection. It's most useful when developers want fine-grained control over conversational timing and are comfortable wiring probability streams into their own agent logic. For those evaluating whether to build or buy turn-taking, the free credits provide a way to measure Sense against their current approach before committing.

---

This is the Markdown form of https://completeaitraining.com/ai-tools/urun/. The whole site is mapped at https://completeaitraining.com/llms.txt.
