Dograh Drops Gemini 3.1 Flash Live Integration: Self-Hosted Voice AI Challenges Vapi and Retell
Dograh launched v1.20, integrating Speech-to-Speech via Gemini 3.1 Flash Live. This collapses the entire STT/LLM/TTS pipeline into a single, more natural connection for voice agents.
Community sentiment focuses on technical capability and vendor independence. Users point to v1.25.0's Pre-call data fetch, allowing connections to CRMs/ERPs via HTTP endpoints. One standout feature is the ability to mix human pre-recorded audio with dynamic TTS for predictable greetings, cited as a cost and quality saver. The core rallying cry remains 'No vendor lock-in,' emphasized by the BSD-2 license and requirement to bring one's own API keys.
The weight of the input confirms Dograh's positioning as a self-hostable, open-source workflow builder directly targeting commercial rivals. The platform's technical stack—FastAPI, Next.js, forked Pipecat, and Langfuse—is presented as robust, with immediate roadmap focus on solving real-time noise separation for live calls.
Key Points
#1The platform enables single-connection voice AI calls.
Harry789 reports v1.20 introduced Speech-to-Speech via Gemini 3.1 Flash Live, collapsing the STT/LLM/TTS pipeline.
#2Integration requires external data sources.
v1.25.0 adds Pre-call data fetch, accepting credentials via HTTP endpoints (POST, API key, bearer).
#3Open source status mandates self-control.
The platform heavily advertises its 'No vendor lock-in' stance, secured by its BSD-2 license.
#4Hybrid audio creation is possible.
Pre-recorded voice mixing lets users combine human recordings for greetings with dynamic TTS to manage costs and improve sound.
#5Advanced monitoring is built into the system.
New features include Post-call QA with sentiment analysis and full call tracing via Langfuse.
#6The major technical hurdle remains noise.
The stated critical roadmap feature is 'Real-time noise separation for live call streams.'
Source Discussions (5)
This report was synthesized from the following Lemmy discussions, ranked by community score.