AI Decision System
Replayable, offline-verifiable records for AI-assisted go/no-go decisions: experiment promotions, typed model answers and human-approved actions.
Connect gRPC, OpenAPI, or MCP. Explore APIs yourself, or let an agent write and run typed scripts. Every request stays visible and inspectable.
Translation pending
Developer toolsTypeScript · GoGitHub @whamOnlineOwner unconfirmed
Have you tried it? One tap is enough. No sign-in needed.
A canvas for your gRPC, MCP and OpenAPI
Taken from the repository README
Loading the full README.
Read the full README ↗Installing Kaja, connecting apps, keeping credentials in variables, writing scripts and pointing an agent at it are all documented on the website.
| Installation | The Mac App Store build, or one docker run with your protos and kaja.json mounted |
| Apps | Connect gRPC, OpenAPI, Twirp and MCP apps, and what each type's block takes |
| Variables | Named values your scripts and your app configuration both read, with secrets kept out of the file |
| Scripts | TypeScript with a typed import for every app you connected, drawing on the canvas |
| Agents | Point your agent at Kaja's MCP server, and watch every run it makes |
Needs Go and Bun; the scripts offer to install them via Homebrew.
scripts/server (add --editable to edit the workspace from the UI)scripts/desktopscripts/ios for the simulator, scripts/ios device for an iPhone (needs full Xcode)scripts/docker(cd ui && bun test) and (cd server && go test ./... -tags development)Every push to main uploads a macOS and iOS build to TestFlight. To ship, run the release workflow from GitHub Actions:
open with patch, minor or major opens a version bump PR. Merge it.ship tags the release and publishes it on GitHub. Run it when a TestFlight build goes to the App Store.