@wolbarg/vercel-ai

Official Vercel AI SDK middleware for Wolbarg shared memory.

Automatically:

1. Recalls relevant memories before generation
2. Injects them as a tagged system message (preserves your existing system prompt)
3. Remembers the conversation after the step completes

Works with generateText, streamText, tool loops, and any AI SDK provider model (OpenAI, Anthropic, Gemini, …).

Requires AI SDK v7+ (LanguageModelV4 middleware).

Install

npm install wolbarg @wolbarg/vercel-ai ai

Peers: wolbarg >= 0.5.4, ai ^7.0.0. Node ≥ 22.

Quick start

import { generateText, wrapLanguageModel } from "ai";
import { openai } from "@ai-sdk/openai";
import { wolbarg, sqlite, openaiEmbedding } from "wolbarg";
import { wolbargMiddleware } from "@wolbarg/vercel-ai";

const memory = wolbarg({
  organization: "my-app",
  storage: sqlite("./memory.db"),
  embedding: openaiEmbedding({
    apiKey: process.env.OPENAI_API_KEY!,
    model: "text-embedding-3-small",
  }),
});
await memory.ready();

const model = wrapLanguageModel({
  model: openai("gpt-4.1-mini"),
  middleware: wolbargMiddleware({
    memory,
    agent: "assistant",
  }),
});

const { text } = await generateText({
  model,
  messages: [{ role: "user", content: "What do I prefer?" }],
});

Or use the convenience helper:

import { createWolbargModel } from "@wolbarg/vercel-ai";

const model = createWolbargModel({
  model: openai("gpt-4.1-mini"),
  memory,
  agent: "assistant",
});

Streaming

import { streamText } from "ai";

const result = streamText({ model, messages });

for await (const chunk of result.textStream) {
  process.stdout.write(chunk);
}

Remember starts when the model emits finish (not only on stream flush), so cancel-after-finish still persists memory. Mid-generation cancel before finish does not remember incomplete turns.

Tools / multi-step

Default remember: "final-step" skips intermediate finishReason.unified === "tool-calls" steps. Recall still runs every step.

Per-request metadata

await generateText({
  model,
  messages,
  providerOptions: {
    wolbarg: {
      sessionId: "s1",
      userId: "u1",
      tags: ["support"],
    },
  },
});

Behavior notes

| Topic | Behavior | | --- | --- | | Recall / remember failures | Soft-fail — never crash model generation (onError / onTelemetry) | | Tools / multi-step | Default remember: "final-step" skips tool-call steps | | Streaming | Remember on finish (+ flush backup); cancel-after-finish is safe | | Multimodal user messages | Text + [attachment:mediaType:filename] placeholders in recall query | | Injected memory | Tagged via providerOptions.wolbarg.injectedMemory + content marker — never re-stored | | Prompt mutation | Caller arrays are never mutated — prompts are cloned |

Docs

https://wolbarg.com/docs/integrations/vercel-ai

License

MIT