@tekmemo/ai-sdk

@tekmemo/ai-sdk

         

Purpose

@tekmemo/ai-sdk makes TekMemo usable as a Vercel AI SDK tool with minimal glue code. It exposes helpers for:

• creating an AI SDK-compatible memory tool
• defining a ready-to-spread tools object
• building a memory-aware system prompt
• using local, cloud, or hybrid TekMemo runtimes behind the same API
• enforcing project, user, conversation, and participant scope boundaries before memory is returned to the model
• generating AgentFS session instructions for Codex, Claude Code, and AI SDK agents

AgentFS session instructions

Use this when an AI SDK-powered agent is working inside an AgentFS session created by @tekmemo/agentfs:

import { createTekMemoAgentSession } from "@tekmemo/agentfs";
import { buildAgentSessionInstructions } from "@tekmemo/ai-sdk";

const session = createTekMemoAgentSession({
  client: agentfsClient,
  memory: tekmemoStore,
  task: "Refactor auth middleware",
});

await session.prepare();

const system = buildAgentSessionInstructions({
  sessionId: session.sessionId,
  task: "Refactor auth middleware",
  paths: session.paths,
});

Install

npm install @tekmemo/ai-sdk ai tekmemo @tekmemo/fs

For hosted memory, also install @tekmemo/cloud-client and create a cloud runtime there.

Plug-and-play AI SDK usage

import { generateText } from "ai";
import { openai } from "@ai-sdk/openai";
import { createNodeFsMemoryStore } from "@tekmemo/fs";
import {
  buildTekMemoSystemPrompt,
  createLocalAiSdkRuntime,
  defineTekMemoTools,
} from "@tekmemo/ai-sdk";

const workspace = createNodeFsMemoryStore({
  rootDir: process.cwd(),
  createRoot: true,
  missingFileBehavior: "empty",
});

const access = {
  projectId: "my-project",
  userId: "user_123",
  conversationId: "thread_456",
  actorId: "assistant:web",
};

const runtime = createLocalAiSdkRuntime({ workspace, access });
const { system } = await buildTekMemoSystemPrompt({
  runtime,
  access,
  query: "What should I remember before answering this request?",
  system: "You are a helpful product engineering assistant.",
});

const result = await generateText({
  model: openai("gpt-4.1-mini"),
  system,
  prompt: "Summarize the current implementation risks.",
  tools: defineTekMemoTools({
    runtime,
    access,
    allowWrites: true,
    allowCoreUpdates: false,
  }),
});

console.log(result.text);

The exported tool name is tekmemo_memory. To control the tool key yourself, use createTekMemoTool() directly:

import { createTekMemoTool } from "@tekmemo/ai-sdk";

const tools = {
  memory: createTekMemoTool({ runtime, access, allowWrites: true }),
};

Local convenience helpers

For local file-backed apps, createLocalTekMemoTool() and defineLocalTekMemoTools() create the local runtime and AI SDK tool in one call.

import { defineLocalTekMemoTools } from "@tekmemo/ai-sdk";

const tools = defineLocalTekMemoTools({
  workspace,
  access: { projectId: "my-project", userId: "user_123" },
  allowWrites: true,
});

Use the explicit runtime + defineTekMemoTools() form when you also need to call buildTekMemoSystemPrompt() with the same runtime.

Safety defaults

Writes are disabled unless allowWrites: true is passed. Core memory updates are disabled unless allowCoreUpdates: true is passed. Indexing is disabled unless allowIndexing: true is passed. Likely API keys and private keys are rejected by default unless allowSecrets: true is intentionally enabled.

Scope filtering is applied using the access object. User memory requires userId, conversation memory requires conversationId, and participant-shared memory requires participantIds.

Boundary

This package owns the AI SDK integration layer only. It does not own TekMemo Cloud billing, dashboards, tenancy, hosted database storage, or provider secrets.

For hosted memory, use @tekmemo/cloud-client. For local file-backed memory, use tekmemo with @tekmemo/fs. For MCP tools, use @tekmemo/mcp-server.

License

MIT.