Mahimai Labs builds open-source infrastructure for voice AI. Cost tracking, multi-agent runtimes, long-term memory and open reference data for teams shipping voice agents on LiveKit and Pipecat.
We are the open-source side of Mahimai AI. The tools here come out of seven years of putting voice systems into production for enterprises, founders and public-sector teams, and they are used every day by the builders who run them.
12 open-source projects · 109 stars · 7 packages on PyPI| Project | What it does | Latest | Stars |
|---|---|---|---|
| voicegateway | Cost tracking, observability and inference routing for voice agents on LiveKit, Pipecat and OpenRTC. | 0.26.1 Sep 2026 |
★ 51 |
| openrtc-runtime | Runs many LiveKit voice agents in one Python worker, sharing heavy models instead of loading them per process. | 0.20.1 Sep 2026 |
★ 11 |
| livekit-plugins-voicemem | Long-term memory for LiveKit voice agents, backed by PostgreSQL and pgvector. Remembers both what the caller said and what they are like. | 0.2.2 Sep 2026 |
★ 2 |
| shipvoice | Full-stack LiveKit voice agent starter: a Python voice worker, a FastAPI token server and a React frontend, each deployable on its own. | updated Aug 2026 | ★ 30 |
| envoic | Finds and reports the Python virtual environments on a machine, so they stop piling up. | 0.3.1 Jul 2026 |
★ 6 |
| Project | What it does | Latest | Stars |
|---|---|---|---|
| voice-prices prices.mahimai.ca |
Open price database for voice AI APIs: STT, LLM, TTS, speech-to-speech and VAD. | 0.11.0 Sep 2026 |
★ 4 |
| voice-ai-skills skills.mahimai.ca |
Vendor-neutral Agent Skills for people who build voice agents. Read by Claude Code, Cursor, Codex, Copilot, Gemini CLI. | updated Sep 2026 | ★ 2 |
| voice-latency | A per-hop latency budget for voice agents: cascaded and realtime (speech-to-speech), plus framework overhead. The numbers are the point, PR better ones. | 0.1.1 Aug 2026 |
★ 1 |
| handset-bench | Benchmark text-to-speech on what survives a G.711 telephone line: 8kHz, mu-law, 300-3400 Hz, packet loss | 0.1.0 Aug 2026 |
★ 0 |
| Project | What it does | Latest | Stars |
|---|---|---|---|
| gpt-live-voice-agent | OpenAI GPT-Live full-duplex voice agent on LiveKit, side by side with a classic STT-LLM-TTS cascade. Run both, interrupt both, see what breaks. | updated Sep 2026 | ★ 1 |
| voicemem-demo | A LiveKit voice agent that remembers you between calls, in ~100 lines. Built on livekit-plugins-voicemem. | updated Sep 2026 | ★ 1 |
| voicegateway-examples | Example projects that run on VoiceGateway. | updated Jun 2026 | ★ 0 |
|
How your voice travels through WebRTC: Voice AI agents edition Your smoke run is lying to you Building a phone line in PyTorch The normaliser that scored itself More on mahimai.ca/blog |
docs(readme): put the badges and cover in one palette, with a light variant refactor(security): remove the Wave 0 contracts for planned telemetry work fix(tests): wait on conditions instead of fixed sleeps ci: run the test suite in parallel with pytest-xdist test: table-driven loadtest tests and a snapshot for the diagnostics report layout fix(storage): take the SQLite write lock up front for call-store writes Contributions are welcome in every repository. |
Mahimai AI takes voice agents from prototype to production: architecture, latency and cost, telephony, and evaluation. mahimai.ca · Book a call
If our open-source work saves your team time, you can support it through GitHub Sponsors.
This page rebuilds itself every six hours. How this works
