TLDR; You don't need to pay OpenAI to log all your ideas and sell them to your competitors... you can just have a local research assistant without any middlemen.

Everyone and their mom in law seems to be super amazed by ChatGPT Deep Research.

To prove that this isn't magic, I pulled jan-nano - a model refined for MCP on top of Qwen3 - with ollama, and connected it with the ddg-search mcp (because that doesn't KYC.) I then hung both in fast-agent:

fastagent.config.yaml:

mcp:
    servers:
        search:
            command: "npx"
            args: ["-y", "@oevortex/ddg_search@latest"]

agent.py:

import asyncio
from mcp_agent.core.fastagent import FastAgent

# Create the application
fast = FastAgent("fast-agent")

@fast.agent(name="researcher", servers=["search"], use_history=False, model="generic.jan-nano:latest")

async def main():
    async with fast.run() as agent:
        await agent.interactive()

if __name__ == "__main__":
    asyncio.run(main())

I then took the last problem a consultant friend told me he was going to have a 4h explorative meeting about: how to run a docker container orchestrator for a lightweight deployment (without k8s).

╭──────────────────────────────────────────────────────── (researcher) [USER] ─╮
│                                                                              │
│  Research what the best orchestrator for docker containers is in             │
│  lightweight but diverse deployments spanning at most 3 host servers.        │
│                                                                              │
╰─ jan-nano:latest turn 1 ─────────────────────────────────────────────────────╯

the agent does the <think></think> thing and then:

╭─ [TOOL CALL] ────────────────────────────────────────────────────────────────╮
│                                                                              │
│  {"numResults":5,"query":"best Docker orchestrator for lightweight           │
│  deployments across 3 host servers"}                                         │
│                                                                              │
╰─ [web] [fetch] [url] [felo]  ────────────────────────────────────────────────╯


╭────────────────────────────────────────────────────────────── [TOOL RESULT] ─╮
│                                                                              │
│  meta=None content=[TextContent(type='text', text='1. [16 Best Container     │
│  Orchestration Tools and Services -                                          │
│  DevOpsCube](https://devopscube.com/docker-container-clustering-tools/)\n    │
│  In this 2025 updated guide, I have covered the best 16 container            │
│                                                                              │
╰──────────────────────────────────────────────────────────────────────────────╯

and then it spits out a massive response block with a recommendation:

╭─ [ASSISTANT] (researcher) ───────────────────────────────────────────────────╮
│  [..]                                                                        │
│  **Recommendation**: Use Docker Swarm for its simplicity, integration, and   │
│  efficiency in small-scale, diverse deployments.                             │
│                                                                              │
╰─ [search]  ──────────────────────────────────────────────────────────────────╯

I still think it doesn't do enough research so I'm going to tune this a bit, but... sovereign "deep research" for just the electricity cost on your old Macbook is here. Took about 2 minutes on my 2020 M1.