Open source · For developers

The tools we built, open to the people building the same things

A translation-pipeline benchmark harness. A reference LiveKit agent. An independent latency canary. All Apache 2.0 or MIT, documented, pinned-version.

We benefited from open-source LiveKit, Opus, and WebRTC, this is the loop closed. If you're building something that touches live call translation, these are the pieces we would have wanted.

The repos

Short tool, clear licence, no telemetry. Build time < 2 minutes on a laptop.

  • RoadmapApache-2.0

    chromeping/call-translation-benchmark

    Measure latency and accuracy of call-translation pipelines

    The harness we use internally to score STT, MT, and TTS stages on real audio samples. Fifty canonical call scenarios (Spanish ayuntamiento, Japanese hotel, US bank), two language pairs per sample, and a dashboard that plots per-stage latency over time. Runs on any VPS; no ChromePing account required.

    TypeScript, Deepgram, Google Translate, Cartesia

    Watch the repoNot public yet, star the org to be notified
  • RoadmapApache-2.0

    chromeping/livekit-translation-agent

    Reference LiveKit agent for live call translation

    A minimal, documented LiveKit agent that forwards call audio through an STT provider, a translation model, and a TTS provider, the same path ChromePing uses in production, with the proprietary bits abstracted behind an interface. Pair with your own carrier credentials to run end-to-end translated calls.

    TypeScript, LiveKit Agents, Node 22

    Watch the repoNot public yet, star the org to be notified
  • RoadmapMIT

    chromeping/latency-canary

    Hourly end-to-end latency measurement for translated calls

    A small canary job that places a translated call once an hour on every supported language pair and records the five-stage latency breakdown. Useful if you're running a production translation pipeline and want an independent third-party measurement that doesn't depend on your own provider.

    Python, Cloudflare Workers, D1

    Watch the repoNot public yet, star the org to be notified

Why we’re publishing these

Three reasons, in descending order of self-interest. First, benchmark harnesses are the single most useful thing a company can give a developer community working on the same problem, a shared yardstick beats competing anecdotes. Second, the reference LiveKit agent documents our own stack choices in a form that outlasts internal docs. Third, open-source presence is a legitimate SEO and GEO lever: generative-engine citers index GitHub heavily, and a well-starred repo earns inbound links the ad budget never could.

What we would like in return

  • Issues, honestly filed. If something is wrong, open an issue. We read every one.
  • Benchmarks on your own hardware. If you run the harness on your own calls and find different numbers than ours, publish them. Multiple independent measurements is how the whole field improves.
  • A star, if it helps you. Stars are a cheap trust signal and they drive the GitHub trending list, which is the single biggest discovery surface for infrastructure code.

Not in the open-source plan

The ChromePing dialer itself, the billing surface, the account system, and the content-translation pipeline internals stay closed-source. Those are the product. The infrastructure adjacent to the product is where open-source makes sense; the product itself is where the business is. See the competitive comparison for how we think about that split.

Working on live call translation? Fork what we published.

Everything on this page is designed to be useful even if you never become a ChromePing customer. If the harness or the agent save you a day of work, we're even.

Browse the GitHub org

Apache 2.0 and MIT licensed. Not for emergency calls.