@nyrra/foundry-ai

Use Vercel AI SDK language models, OpenAI embeddings, and realtime audio through Palantir Foundry's proxy endpoints.

Install

Install only the provider peers you use:

pnpm add @nyrra/foundry-ai ai@7 @ai-sdk/openai@4

Use @ai-sdk/anthropic for Claude and @ai-sdk/google for Gemini. Use @ai-sdk/xai@4 for Grok; third-party models use @ai-sdk/openai. Use stable AI SDK 7 with provider v4 for new integrations. AI SDK 6/provider v3 and the late SDK 7 beta line from 7.0.0-beta.187 remain compatible, but the current live survey uses stable SDK 7.

Configure and call a model

Set FOUNDRY_URL and FOUNDRY_TOKEN in your server environment. Optional variables are FOUNDRY_ATTRIBUTION_RID, FOUNDRY_TRACE_PARENT, and FOUNDRY_TRACE_STATE. Keep server tokens out of browser code.

import { loadFoundryConfig } from '@nyrra/foundry-ai';
import { createFoundryOpenAI } from '@nyrra/foundry-ai/openai';
import { generateText } from 'ai';

const openai = createFoundryOpenAI(loadFoundryConfig());
const { text } = await generateText({
  model: openai('gpt-5-mini'),
  prompt: 'Reply in one sentence.',
});
console.log(text);

For explicit configuration, pass { foundryUrl, token } to the factory. Other factories are createFoundryAnthropic, createFoundryGoogle, createFoundryXai, and createFoundryThirdParty, imported from their respective anthropic, google, xai, and third-party subpaths.

Models and test status covers model IDs, routing, known failures, and dated live results. Catalog membership alone does not establish feature support. CI verifies SDK compatibility offline; live Foundry probes are manual.

xAI / Grok

Install @ai-sdk/xai@4 with AI SDK 7 (or @ai-sdk/xai@3 with SDK 6). The xAI subpath does not require the OpenAI peer.

import { createFoundryXai } from '@nyrra/foundry-ai/xai';

const xai = createFoundryXai(loadFoundryConfig());
const model = xai('grok-4-6'); // Defaults to the xAI Responses route.
// Explicit routes: xai.responses('grok-4-6') or xai.chat('grok-4-6').

Grok uses model identity foundry-xai, catalog provider xai, registry prefix xai:, and providerOptions.xai. Both endpoints are beta. See xAI setup and migration and the current Grok results for verified capabilities and remaining gaps.

Embeddings and multiple providers

import { embed } from 'ai';

const { embedding } = await embed({
  model: openai.embeddingModel('text-embedding-3-small'),
  value: 'Text to embed',
});

Compose multiple providers with AI SDK createProviderRegistry; this package exports no registry wrapper. See the runnable examples for registry and tool workflows.

Server environment configuration is the validated runtime path. Palantir TSv1/TSv2 functions and PlatformClient fetch integration have not been verified end to end.

Realtime (AI SDK 7 only)

Use createFoundryRealtime and createFoundryRealtimeSetup from @nyrra/foundry-ai/realtime with gpt-realtime, gpt-realtime-1.5, or gpt-realtime-2. Install ai@7 and @ai-sdk/openai@4; this subpath uses the SDK's experimental realtime interface and does not support SDK 6.

Foundry authenticates WebSockets with the current user's access token. It does not use OpenAI's client-secret endpoint. See realtime setup and live results for the browser flow and the manual test command.

Agent skill

npx skills add https://github.com/shpitdev/foundry-ai --skill foundry-ai-provider

The package also ships the skill at skills/foundry-ai-provider/SKILL.md for TanStack Intent discovery.

Copyright 2026 SHPIT LLC. Apache-2.0.