- CollectionVoice & Phone
- Classificationagent-native
- StatusOpen source · early-stage but rapid 2026 iteration
- MCPUnavailable
- Agent SkillUnavailable
- InterfacesAPI · documentation
What it does
Stimm is an open-source orchestrator for real-time voice agents over WebRTC. Its design center is ultra-low latency with two distinguishing primitives:
- Dual-agent architecture — a small, fast agent handles immediate verbal feedback (filler / acknowledgment) while a larger, slower agent handles the substantive reasoning. The pipeline coordinates the two so the user never feels a long silence.
- Optimistic VUI — speak interim feedback before the full LLM response is finalized, cancel or amend if reasoning diverges.
Operators bring their own STT, LLM, and TTS providers; Stimm orchestrates the WebRTC media plane and the dual-agent control plane. It is an open-source contender to LiveKit Agents and the closed-source Vapi / Retell, with a tighter focus on perceived-latency engineering.
Agent-facing interfaces
- Catalog protocols: API, documentation
- MCP: ⚠️ Not yet published as a first-party MCP server.
- Agent Skill: ⚠️ Not yet published.
- Onboarding type:
documentation
How agents onboard
git clone https://github.com/stimm-ai/stimm
# follow README to install + run the orchestrator
# then connect your STT / LLM / TTS providers and a WebRTC client
See the full onboarding section in the dossier.
Protocol surface
| Interface | Detail |
|---|---|
| WebRTC | Real-time audio media plane |
| Orchestrator API | Pipeline configuration, session lifecycle |
| Provider adapters | STT, LLM, TTS — pluggable |