Skip to content

How a writer made ChatGPT a creative partner that remembers the whole story

Jim Johnston is writing a large story. It has a lot of connected material: characters, history, organizations, rules, locations, timelines and relationships.

For a while, ChatGPT kept up. Then the project grew, and important details started falling out of the conversation. The AI contradicted earlier material. It forgot what had already been established. Worst of all, it sometimes confidently invented something that was never part of the story.

He tried the usual fixes: uploading files, and explaining things again at the start of each chat.

"Too much time went on uploading files and explaining things we'd already worked out."

He tested it before he paid

ChatGPT suggested Hjarni while Jim was looking for a real external memory for the project. He didn't take that on trust.

He loaded real material from the story into Hjarni and tried to catch ChatGPT out. He asked about details buried deep in the material. He asked whether a newly written chapter actually followed the events and character development of the chapters before it. ChatGPT went into Hjarni, found the right notes, and answered from the source instead of guessing.

Once it answered from his material reliably, the decision was easy. He upgraded to Pro straight away: of everything he had tried, it was the best fit for what he was building.

Consistency got him started. It isn't the best part.

When Jim has an idea for where a character or storyline could go, he doesn't ask ChatGPT to develop it in isolation.

"I have it check Hjarni first for the established history, personality, relationships, previous decisions, unresolved threads and future ideas connected to it. That changes the quality of the discussion completely."

Because ChatGPT works from the actual history of the project, its suggestions stop being generic. It can pick up a character trait, an earlier event, an unresolved relationship, or a small detail from much earlier in the story, and show where it could logically lead.

Jim stays the author. Sometimes his original idea wins. Sometimes one of ChatGPT's alternatives is stronger. Quite often the best result combines parts of several. Once they settle on a direction, ChatGPT saves that decision back into Hjarni.

His workflow, in his own words:

"I have an idea → Hjarni provides the established context → ChatGPT explores and challenges the idea → we find stronger possibilities → the result goes back into Hjarni."

How Jim works

  1. The story's memory

    Hjarni

    • History
    • Relationships
    • Earlier decisions
    • Open threads
  2. 01Jimhas an idea for a character or storyline
  3. 02 readsHjarniChatGPT reads what's already established
  4. 03ChatGPTexplores and challenges the idea from what's established
  5. 04Jim + ChatGPTsettle on the strongest direction
  6. 05 savesHjarniChatGPT saves the decision back
  7. Next ideaone more decision on record

The memory grows with the story

Every decision that goes back in becomes material for the next conversation. The notes stop being a static archive and become a living reference for the story.

"The more the project grows, the more valuable that becomes."

The memory doesn't belong to one assistant either.

"I've given both ChatGPT and Claude access to it so they can check and update the same notes."

For Jim, that is where Hjarni stopped being a notes app.

"It gives ChatGPT the depth of context needed to have a genuinely useful creative discussion with me, and the decisions that come out of those discussions become part of the material available for future work."

Set it up for your own story

  1. Connect Hjarni to ChatGPT or Claude. It takes a few minutes, and both can use the same notes.
  2. Give the story a home. A folder for characters with one note each, a folder for the world, and a timeline note. Add what's already established.
  3. Tell your assistant how to use it. Add a line to ChatGPT's or Claude's own instructions (custom instructions or project instructions), such as: "Before we develop a character or plot idea, check the relevant notes in Hjarni. When we agree on a direction, save the decision back." ChatGPT and Claude only search when they decide to, so this line is what makes it a habit. Where to add it, in ChatGPT or Claude.
  4. Test it the way Jim did. Ask about a detail buried deep in your notes, and see whether the answer comes from your story or from guesswork.

More setups for writers are on Hjarni for Writers.

Give your story a memory

Hjarni is a knowledge base your AI can read and write. Your notes stay plain Markdown you can open, edit and export. It works with ChatGPT and Claude, and it's free to start.

Start your story's memory

Thanks to Jim Johnston for sharing his workflow. Quotes are from his emails, used with permission.

Give your AI a memory. Free.

Connect Claude or ChatGPT to notes they can actually read and write.

Get started free

Give your AI a memory. Free.

Start here

Write once. You both remember.

Free to start. No credit card required.

5 out of 5 from 9 reviews 5,900+ notes a day read or updated by AI

Works with Claude and ChatGPT today.