@agentified/ai-sdk

Vercel AI SDK adapter for Agentified — wraps SDK classes with AI SDK-typed shells via composition. Tools are exposed as AI SDK tool() objects for direct use with generateText/streamText.

Install

npm install @agentified/ai-sdk agentified

Peer dependencies: ai >= 6.0.0, zod >= 3.0.0

Quick Start

import { Agentified } from "agentified";
import { aiSdk } from "@agentified/ai-sdk";
import { openai } from "@ai-sdk/openai";
import { generateText, stepCountIs } from "ai";

const ag = new Agentified().adaptTo(aiSdk());
await ag.connect("http://localhost:9119");

const instance = await ag.dataset("my-agent").register({
  tools: [
    { name: "get_weather", description: "Get weather", parameters: { type: "object", properties: { city: { type: "string" } }, required: ["city"] }, handler: async (args) => ({ temp: 22 }) },
  ],
});

const session = instance.session("chat-1");

// All tools passed upfront, prepareStep controls activeTools per step
const result = await generateText({
  model: openai("gpt-4o"),
  tools: session.tools,
  prepareStep: session.prepareStep,
  stopWhen: stepCountIs(10),
  prompt: "Use agentified_discover to find tools, then call them.",
});

// Persist final step's messages (prepareStep only flushes previous steps)
await session.flushMessages(result.steps);

With Context Builder

const ctx = await session.context
  .tools({ agentified_discover: session.discoverTool })
  .messages({ strategy: "recent", maxTokens: 4000 })
  .recall({ tools: true })
  .assemble();

const result = await generateText({
  model: openai("gpt-4o"),
  tools: ctx.tools,
  prepareStep: ctx.prepareStep,
  stopWhen: stepCountIs(10),
  messages: ctx.messages.map(m => ({ role: m.role as any, content: m.content })),
});

await ctx.flushMessages(result.steps);

Key Differences from Mastra Adapter

| Concern | Mastra (@agentified/mastra) | AI SDK (@agentified/ai-sdk) | |---|---|---| | Tool delivery | prepareStep returns { tools } per step | tools property upfront; prepareStep returns { activeTools: string[] } | | Tool creation | createTool() from @mastra/core | tool() from ai | | Final step | Handled by Mastra | Call flushMessages(result.steps) after generateText | | Streaming | streamSSE() export | Use streamText() from ai directly |

Adapter Pattern

Agentified.adaptTo(aiSdk()) → AiSdkAgentified
  └─ .dataset(name) → AiSdkDatasetRef
       └─ .register({ tools }) → AiSdkInstance
            ├─ .tools            — Record<string, CoreTool> (all tools upfront)
            ├─ .discoverTool     — AI SDK tool()
            ├─ .prepareStep      — returns { activeTools: string[] }
            ├─ .session(id)      → AiSdkSession
            │    ├─ .tools             — includes discover + getMessages + backend
            │    ├─ .discoverTool      — AI SDK tool()
            │    ├─ .getMessagesTool   — AI SDK tool()
            │    ├─ .prepareStep       — returns { activeTools: string[] }
            │    ├─ .flushMessages(steps) — persist final step
            │    ├─ .context           → AiSdkContextBuilder
            │    │    ├─ .tools() / .messages() / .recall() / .limitTokens()
            │    │    └─ .assemble()   → AiSdkAssembledContext
            │    │         ├─ .tools          — explicit + discovered
            │    │         ├─ .prepareStep    — returns { activeTools }
            │    │         └─ .flushMessages  — persist final step
            │    ├─ .conversation (SDK passthrough)
            │    └─ .getMessages / .updateConversation
            └─ .namespace(id)    → AiSdkNamespace
                 └─ .session(id) → AiSdkSession

API Reference

aiSdk()

Returns an adapter object for Agentified.adaptTo().

AiSdkInstance

Wraps SDK Instance. Exposes tools (all registered tools as AI SDK objects) and prepareStep (returns { activeTools }).

AiSdkSession

Wraps SDK Session. Exposes tools, prepareStep, flushMessages, and context builder. discoverTool and getMessagesTool are AI SDK tool() results.

flushMessages(steps)

Call after generateText/streamText completes to persist the final step's messages. The SDK's prepareStep persists messages from previous steps, but the last step has no subsequent prepareStep call.

jsonSchemaToZod(schema)

Converts a JSON Schema object to a Zod schema. Used internally to hydrate AI SDK tools.

Links

• Root README
• Documentation
• Architecture
• TypeScript SDK

License

MIT