Should I use Ownhand or ChatGPT memory and Projects to write in my voice?
If you draft in ChatGPT and mostly need it to remember who you are and what you are working on, memory plus a Project that holds a few of your real messages is free and often enough; add Ownhand when you want each draft rewritten from your own samples for a specific occasion and reader, and changed by the exact text you sent, with every learned rule visible and reversible.
ChatGPT facts come from OpenAI's help center and developer docs as of 4 October 2026. I run Ownhand, so read my take with that in mind and check the links.
Custom instructions are covered on Ownhand vs custom instructions. This page is about the two ChatGPT features people usually try next: memory, which carries what ChatGPT knows about you between chats, and Projects, where you can park samples of your writing.
What memory and Projects do well
Memory is ChatGPT's running picture of you. Depending on your plan and region, OpenAI says it can draw on past chats, saved memories, your custom instructions, files in your Library and content from connected apps such as Gmail. Tell it your accountant is Sam and that you sign emails with your first name, and next week it still knows both.
A Project groups chats, files and instructions. Its instructions override your global custom instructions inside that project, so a project called "Client emails" can carry rules that would clutter every other chat. You can upload files, paste text, add a Google Drive folder or a Slack channel as a source, and save a good response back to the project for reuse.
Both come with your plan at no extra charge. If your real problem is that ChatGPT forgets your product names, your team and the deal you are in the middle of, these fix it. Ownhand stores your samples, your rules and notes on the people you write to. It does not carry project context from one chat to the next, so it is the wrong home for that.
Getting the most from a writing project
File limits depend on your plan, and you can upload 10 files at a time:
| Plan | Files per project |
|---|---|
| Free | 5 |
| Go, Plus | 25 |
| Pro, Business, Enterprise, Edu | 40 |
Five files sounds tight, but one document can hold twenty short messages. Put each sample under a label like "[Slack DM to my manager]" or "[first email to a prospect]" so ChatGPT can tell a two-line DM from a client email.
Be careful with the Slack source. A channel link pulls in everyone's messages, so the project learns how the channel writes. Paste your own messages instead.
Set the project to project-only memory. Its chats then stop drawing on your saved memories and on chats outside the project. Under default memory, your saved memories and chats from the rest of ChatGPT can feed into the drafts you make there.
Where they stop
Memory chooses what to keep. OpenAI's own wording is that memory does not retain every detail, and ChatGPT decides what is relevant to each response. The memory summary you can read is a high-level view that may leave things out. You can correct it, or tell it not to mention something again. You cannot tell it to study how you write to clients and hold on to exactly that.
Project memory has no list. OpenAI's FAQ says plainly that you can't see a list of project memories. ChatGPT may use other conversations in the same project as context, and the way to stop it is to delete the chat or move it out. So the three drafts you threw away last Tuesday may still be shaping today's.
Nothing learns from the sent version. You take a draft from ChatGPT, rewrite half of it in Gmail and hit send. Even when memory can read Gmail, OpenAI describes it as context for answers. No step in its docs turns the gap between the draft and your sent email into a rule for next time.
One project, one voice. A project gives the same samples to every message you draft in it. You can make as many projects as you like, one per kind of message, but then you have to open the right one each time.
Shared projects aim for a house style. Among other uses, OpenAI suggests them for keeping tone and style consistent across contributors. That suits a team blog. It works against sounding like one particular person, and a shared project can't see any member's own memories.
What Ownhand keeps instead
| ChatGPT memory | ChatGPT Project | Ownhand | |
|---|---|---|---|
| What it holds | Details and preferences from your chats and other sources | Files, chats and project instructions | Your samples, rules learned from what you sent, and the people you write to |
| Can you read it | A summary that may omit details | The files yes, the project memory no | Every rule, marked proposed, active, pinned or rejected |
| Undo | Correct, hide or delete items | Delete or move a chat | Reject a rule, or roll the Hand back to an earlier version |
| Learns from what you sent | No such step described | No such step described | Yes, from the exact text your agent reports |
| Kinds of message | The same context for every kind of message | One set of samples per project | 12 occasions, plus up to 20 of your own |
| Per reader | Whatever it picks up | None | Rules that apply to one person, matched by email or name |
| Where it works | ChatGPT | ChatGPT | Any MCP client, including ChatGPT developer mode |
| Cost | Included | Included | $1 free, then about half a cent per rewrite |
The per-reader row is the one memory can't copy. Say you sign off "J" with Dana, your cofounder, and use your full name with clients. You change the sign-off on a draft to Dana, send it, and your agent reports what went out. Ownhand proposes a rule for Dana alone, and it goes active the second time the same edit shows up. If you give a reason in your own words, like "too stiff for Dana", it goes active straight away. Learning takes a minute or two in the background. The full mechanics are in how Ownhand learns from what you send.
Replies work differently too. Your agent passes up to 30 recent messages from the thread. Ownhand measures how you write in that thread, things like casing, length and sign-offs, and checks the rewrite against it. The thread is never stored.
Same draft, two writers
These examples are real output from the live API. Each AI draft was rewritten by a demo Hand: a fictional writer (an engineer, a founder, an ops lead, a designer or a consultant) built from a handful of samples.
AI draft: Please accept my sincere apologies for the error in last month's invoice. We have issued a corrected invoice for $1,240 and will ensure this does not happen again.
Demo Hand, ops: Sorry about the error in last month's invoice. Corrected one is out for $1,240.
AI draft: Please accept my sincere apologies for the error in last month's invoice. We have issued a corrected invoice for $1,240 and will ensure this does not happen again.
Demo Hand, consultant: My apologies for the error in last month's invoice. I've sent the corrected invoice for $1,240. It won't happen again.
A memory note saying "keep apologies short" would fit both of these. The difference is "Sorry about" against "My apologies", "Corrected one is out" against "I've sent", and whether the promise survives. Both keep the $1,240. That level of habit comes from samples and from the edits a writer makes, which is what a Hand is built on.
Using them together
They don't compete for the same job. Leave memory on so ChatGPT knows your world. Keep a project for long-running work. Then add Ownhand as a developer mode app, which OpenAI offers on Plus, Pro, Business, Enterprise and Education accounts on the web, so drafts that go out under your name get rewritten from your Hand. Setup steps are in how to make ChatGPT write in your voice. The same Hand follows you into Claude, Cursor and other MCP clients.
Stay with memory and Projects if
- You are on the free ChatGPT plan, where developer mode isn't available. A project with one well-labeled sample file is your best free setup.
- Your complaint is that ChatGPT forgets context, and the tone is already close enough.
- You draft one kind of message, in ChatGPT only, and rarely edit before sending.
- A team wants one shared tone. A shared project is built for that, and Ownhand builds one writer's Hand at a time.
Add Ownhand if
- You keep rewriting ChatGPT's drafts before they go out, and the same fixes come back every week.
- You write to the same people often and want each of them to get the version of you they know.
- You reply in threads a lot and want replies that match the people in them.
- You draft in more than one tool and want one voice across all of them.
Follow-up questions
Should a writing project in ChatGPT use default or project-only memory?
For a project that holds your samples, project-only memory is the cleaner choice. Its chats stop referencing your saved memories and chats from outside the project, so your casual back and forth with ChatGPT stays out of the drafts. You pick it in Project settings under Memory, and OpenAI says the change can take a few hours. Shared projects always use project-only memory.
Can ChatGPT memory read the emails I send?
On some plans and regions, yes. OpenAI lists content from connected apps such as Gmail among the sources memory can draw on, and ChatGPT decides what is relevant to a given response. OpenAI describes that as context for personalizing answers. Its docs describe no step that compares a draft ChatGPT gave you with the version you sent.
Does Ownhand see my ChatGPT memory or my Projects?
No. Ownhand only receives what ChatGPT passes in a tool call: the draft, the occasion, and any thread or background material the agent includes. Your memory and project files stay with OpenAI.
Is my writing used to train models in either one?
On personal ChatGPT accounts, OpenAI may use chats and remembered information to improve its models while Improve the model for everyone is on, and you can turn that off in Data controls. Business, Enterprise and Edu workspaces are excluded by default. Ownhand does not train its own models on your writing. The model provider it runs on may use what it receives to improve its models, as the safety page explains.
Sources
Checked on 4 October 2026. If something here is out of date, email hi@thalientlabs.ai and we will fix it.
- OpenAI Help Center: Memory in ChatGPT: sources memory draws on, the memory summary, saved memories and chat history, temporary chats, model training
- OpenAI Help Center: Projects in ChatGPT: files per plan, Slack and Google Drive sources, project instructions, default and project-only memory, no list of project memories, sharing
- OpenAI Developers: ChatGPT developer mode: Plus, Pro, Business, Enterprise and Education accounts on the web
- Ownhand source code: occasions, people, rule statuses, reply matching and the feedback loop