Thin Palantir Foundry adapters and model catalog for the Vercel AI SDK and Effect AI.
- Install and use the package
- Models and test status
- Runnable examples
- Effect AI adapters (
@shpitdev/foundry-effect-ai) and their models and test status - Release process
mise.toml pins Node.js, pnpm and actionlint.
mise install
pnpm install --frozen-lockfile
pnpm run verifyverify runs release-policy checks, lint, Vitest, typecheck, build, package-content audit, and the Effect package's tree-shaking check through Nx for both packages. CI also tests the packed library with AI SDK 6, SDK 7 stable, and the supported late SDK 7 beta line.
| Test layer | Coverage | CI |
|---|---|---|
| Unit | Config, catalog, adapter behavior | Yes |
| Integration | Offline SDK tool loop, packed-library compatibility | Yes |
| API end-to-end | Manual live Vitest suite against Foundry | No; requires credentials |
| Web end-to-end | None | No |
Run pnpm run test:live for the default live models (it reads FOUNDRY_URL and FOUNDRY_TOKEN from Infisical after infisical login), or LIVE_MODELS=gpt-6-luna pnpm run test:live for one model. LIVE_MODELS=all surveys the catalog. The test status page explains selection and evidence limits.
Provider factories stay thin: config and catalog exports live at the package root, provider clients use separate subpaths, and applications compose registries with AI SDK createProviderRegistry.
Copyright 2026 SHPIT LLC. Apache-2.0. See security reporting.