@agentrail/deep-research

Reusable Deep Research workflow package for Agentrail. It provides a multi-agent research coordinator and routes, but does not replace the general hosted app layer.

Installation

pnpm add @agentrail/deep-research @agentrail/app @agentrail/core

Quick example

import { SessionManager } from "@agentrail/app";
import { createDeepResearchRunRoute } from "@agentrail/deep-research";

const sessionStore = new SessionManager("./.agentrail");

const route = createDeepResearchRunRoute({
  sessionStore,
  runtime: {
    dataDir: "./.agentrail",
    model: {
      provider: "openai",
      modelId: "gpt-4.1-mini",
    },
    searchProvider: "tavily",
    tavilyApiKey: process.env.TAVILY_API_KEY,
  },
});

API

• createDeepResearchRunRoute(options) — mount a blocking deep-research run endpoint.
• createDeepResearchRoute(options) — read deep-research state and artifacts.
• runDeepResearchBlocking(input) / runDeepResearchStreaming(input) — run the workflow programmatically.
• DeepResearchCoordinator — lower-level workflow coordinator.
• createFileSystemDeepResearchStore(dataDir, sessionRef) — persisted state/event store for deep-research runs.

Reference:

• https://agentrail.run/examples/deep-research

Related packages

• @agentrail/app — provides session storage and hosted routes used around the workflow.
• @agentrail/capabilities — provides orchestration, sandbox, and web tools consumed by the workflow.

License

Apache-2.0