Agentdock gives TypeScript applications a focused runtime for model calls, typed tools, approvals, sessions, persistence, streaming, and lifecycle control. LangChain and LangGraph run internally; application code uses the Agentdock API.

✨ What you get

• Agent runtime: one AgentDock class for runs, streams, and resumes.
• Typed tools: validation, authorization, progress, cancellation, and approvals.
• Durable sessions: memory, SQLite, PostgreSQL, MongoDB, and Redis adapters.
• Normalized events: one frontend-friendly contract for text, tools, usage, and interrupts.
• Safe lifecycle: timeouts, cancellation, cleanup, and owned resource management.

🚀 Install

yarn add @agentdock-ai/agentdock @agentdock-ai/models zod

💻 Quick start

import { AgentDock, defineTool } from "@agentdock-ai/agentdock";
import { AgentDockModel } from "@agentdock-ai/models";
import { z } from "zod";

const weather = defineTool({
  name: "get_weather",
  description: "Get the weather for a city.",
  input: z.object({ city: z.string() }),
  run: async ({ city }) => ({ city, forecast: "Sunny" }),
});

const agent = new AgentDock({
  model: AgentDockModel.openAI({ model: "gpt-5.4-mini" }),
});

agent.registerTool(weather);

try {
  const result = await agent.run(
    "What is the weather in Lahore?",
    { userId: "user-123" },
    { sessionId: "session-123" },
  );

  console.log(result.content);
} finally {
  await agent.close();
}

Use agent.stream() when the UI should receive text and tool activity as it arrives:

const { stream, result } = await agent.stream(
  "Summarize my latest order.",
  { userId: "user-123" },
  { sessionId: "session-123" },
);

for await (const event of stream) {
  if (event.type === "message.part.delta" && event.part.type === "text") {
    process.stdout.write(event.part.text);
  }
}

console.log(await result);

🧠 Sessions and approvals

Pass a stable sessionId to continue a conversation. Use a durable checkpoint adapter when sessions must survive restarts or be shared across instances:

yarn add @agentdock-ai/checkpoint-postgres

import { PostgresCheckpoint } from "@agentdock-ai/checkpoint-postgres";

const agent = new AgentDock({
  model,
  checkpoint: new PostgresCheckpoint({
    connectionString: process.env.DATABASE_URL!,
  }),
});

Set requiresApproval: true on a side-effecting tool. Agentdock pauses the run, persists the interrupt, and resumes it with agent.resume() after approval.

📚 Useful APIs

| API | Use it for | | --------------------- | --------------------------------------------------- | | run() | Execute a prompt and receive one result. | | stream() | Consume normalized events while a run is executing. | | resume() | Continue a paused approval run. | | getSession() | Read the current normalized message state. | | getSessionHistory() | Inspect checkpoint-by-checkpoint history. | | deleteSession() | Remove a session’s checkpoint context. | | close() | Stop active work and release owned resources. |

🔗 Related packages

• @agentdock-ai/models — provider configuration.
• @agentdock-ai/contracts — framework-independent events and data types.
• @agentdock-ai/checkpoint — checkpoint contract and memory adapter.

📄 License

MIT. See the repository license.