Build with the MeetKanoe project.

Source repository

The canonical public source is the harishmaddali/meetkanoe repository on GitHub. It uses pnpm workspaces and Turborepo. The web package is a Next.js application; recorder and orchestration packages are Node.js and TypeScript services. Repository documentation explains the current recorder architecture and production considerations.

Recorder architecture

The Google Meet WebRTC recorder is designed as a one-shot process: it receives meeting configuration, joins a call, captures media, sends output to configured storage, and reports completion. The orchestrator creates isolated Kubernetes Jobs for recording requests and exposes a health endpoint plus an authenticated recording endpoint for the Kanoe service. Deployment requires operators to supply their own runtime and storage configuration.

Public product-information API

The stable /api/v1 surface provides read-only JSON for MeetKanoe's public identity, supported platform, beta features, and advertised plans. It requires no authentication because it exposes no user or meeting data. Read the API guide or fetch the OpenAPI 3.1 description. Each documented operation has a unique operation ID and typed response and error schemas.

Separate private application routes support the first-party Kanoe interface for meetings, assistants, categories, and user settings. They require an authenticated account where appropriate and are not part of the public product-information contract. The public v1 API does not start recordings, change accounts, expose transcripts, or provide other writes.

Machine-readable discovery

Agents can start with /llms.txt, enumerate indexable public pages through /sitemap.xml, and request a Markdown representation of a public page with an Accept: text/markdown header. The same canonical URL serves HTML to browsers and Markdown to compatible agents. Unsupported API routes return structured JSON errors with a resolution hint instead of an HTML page.