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 · 114 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. | 0.27.0 Oct 2026 |
★ 53 |
| 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 Oct 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 |
★ 6 |
| 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 | ★ 2 |
| 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 |
|
Build a Real-Time AI Sales Copilot with LiveKit Bilingual Voice Agent: When to Switch Languages Config-Driven Voice Agents: Swap Prompts and Tools Mid-Call Full Duplex vs Cascade Voice Agents: A Blind Test Live QA for Voice Agents: Grade Every Call as It Runs More on mahimai.ca/blog |
test: trim the suite to tests that guard behaviour, and make it 3x faster chore: sanitize dead code and cut unused features before V1 feat(prices): optional FX conversion at set time, with provenance chore(license): adopt open-core license with ee/ Enterprise Edition feat: let a pool skip the multilingual turn detector feat: per-agent start options and a pre-start hook 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
