Research Handoff
Research in ChatGPT, then write the report in Claude Code. Mem carries the context across the seam so you do not re-explain your own work.
Some work starts in one tool and has to finish in another. A common shape: you spend an evening in ChatGPT researching a company, and the investment report itself needs to be written in Claude Code, where your files, templates, and version history live. The seam between the two tools is where context usually dies. You end up pasting fragments, or worse, re-deriving conclusions you already reached.
This walkthrough closes that seam with Mem. You save the ChatGPT research once. Claude Code then recalls it and drafts the report from what you actually concluded, not from scratch.
What You Need
Set up once. Each linked page owns its own install steps and success check, so finish those pages first if either connection is new to you:
- Nowledge Mem installed and running (Installation)
- The browser extension connected to ChatGPT (ChatGPT Web and Desktop, Option 1)
- The Claude Code plugin with the
nmemCLI (Claude Code)
The Walkthrough
Do the research in ChatGPT
Work the way you normally would. Dig into the company's filings, question the growth story, argue with the model about the risks. The research does not need to be shaped for Mem while you do it.
When the thread holds conclusions worth keeping, open the Nowledge Mem Exchange side panel on the ChatGPT page:
- Thread Backup saves the full conversation into Mem as a thread.
- Manual Distill extracts the durable takeaways as memories.
What you should see: in Mem, the conversation appears under Threads with ChatGPT as its source, and the distilled memories show up in your Timeline.
Recall can only find what was saved
If distillation missed a conclusion you care about, add it yourself in the Timeline before moving on. A one-line memory written in your own words is enough.
Confirm the handoff in Mem
Before switching tools, search Mem for a key term from the research. The company name works well. Both the distilled memories and the ChatGPT thread should come back.
This is the success gate for the whole handoff: if Mem search finds it, Claude Code can find it too.
Pick it up in Claude Code
Open a new Claude Code session in the project where the report will live. The plugin loads your Context Bundle or Working Memory at session start, so recent research is often already in view. Then ask for the recall explicitly:
Before we draft anything, search my Mem for the <company> research
I did in ChatGPT. List what we concluded and what stayed open.Claude Code searches your memories and threads and answers from your saved research.
What you should see: the recall reflects your conclusions, with the saved ChatGPT thread as their source. If a detail is missing, ask Claude Code to open the full thread and pull the reasoning behind a specific point.
Write the report
Now hand over the drafting:
Draft the investment report on <company> in reports/<company>.md.
Use the recalled research as the factual base. Where the research
left a question open, keep it marked open instead of filling the
gap with a guess.The draft lands as a real file in your repository, grounded in the research you already did. Mem moves the context; the judgment in the report is still yours to review.
When the draft is where you want it, run /save so the report session is captured too. The next tool that asks about this company will see both the research and the report.
The Handoff Moment
Illustrative example
This exchange is constructed from the shipped behavior described above, with a fictional company. It is not a transcript of a live run. A sanitized transcript from a real walk will replace it.
> Before we draft anything, search my Mem for the Northwind Robotics
> research I did in ChatGPT. List what we concluded and what stayed open.
Running: nmem m search "Northwind Robotics"
I found your research, distilled from the ChatGPT thread
"Northwind Robotics deep dive":
Concluded:
- Revenue quality: most revenue is recurring service contracts;
hardware margin is thin and narrowing.
- The growth story is the warehouse-automation pivot; both named
pilot customers renewed.
- Main risk is customer concentration: a few large accounts carry
the service revenue.
Still open:
- No conclusion yet on how incumbent vendors respond.
- Pilot unit economics were estimated in the research, not sourced.
Want me to open the full ChatGPT thread for the reasoning behind
any of these before we start the draft?Next Steps
- ChatGPT Web and Desktop: the capture side, including importing older ChatGPT history
- Claude Code: the recall side, lifecycle hooks, and session capture
- Never Lose a Session: keep every important conversation saved