--- title: "Edit What Your AI Assistant Remembers About You" url: https://www.insulin.dev/blog/edit-what-your-ai-assistant-remembers/ canonical: https://www.insulin.dev/blog/edit-what-your-ai-assistant-remembers/ type: Blog description: "How to edit what your AI assistant remembers in Insulin: where memories sit, how corrections work, what Clear All removes, and why wording decides recall." --- # Edit What Your AI Assistant Remembers About You > Canonical HTML version: https://www.insulin.dev/blog/edit-what-your-ai-assistant-remembers/ 1. [Home](/) 2. / 3. [Blog](/blog/) 4. / 5. Edit What Your AI Assistant Remembers About You # Edit What Your AI Assistant Remembers About You How to edit what your AI assistant remembers in Insulin: where memories sit, how corrections work, what Clear All removes, and why wording decides recall. ![Sophia Faria](/authors/sophia-faria.jpg) Sophia Faria Sep 19, 2026 ![Edit What Your AI Assistant Remembers About You](/images/blog/edit-what-your-ai-assistant-remembers/hero.png) Explore AI Summary [![](/logos/company/openai.svg)](https://chat.openai.com/?q=Read%20and%20summarize%20https%3A%2F%2Fwww.insulin.dev%2Fblog%2Fedit-what-your-ai-assistant-remembers%2F%2C%20then%20cite%20the%20source.%20Focus%20on%20what%20it%20says%20about%20Trust%2C%20Workspace. "Summarize with ChatGPT")[![](/logos/company/anthropic.svg) ](https://claude.ai/new?q=Read%20and%20summarize%20https%3A%2F%2Fwww.insulin.dev%2Fblog%2Fedit-what-your-ai-assistant-remembers%2F%2C%20then%20cite%20the%20source.%20Focus%20on%20what%20it%20says%20about%20Trust%2C%20Workspace. "Summarize with Claude")[![](/logos/company/gemini.svg)](https://www.google.com/search?udm=50&aep=11&q=Read%20and%20summarize%20https%3A%2F%2Fwww.insulin.dev%2Fblog%2Fedit-what-your-ai-assistant-remembers%2F%2C%20then%20cite%20the%20source.%20Focus%20on%20what%20it%20says%20about%20Trust%2C%20Workspace. "Summarize with Gemini")[](https://www.perplexity.ai/search/new?q=Read%20and%20summarize%20https%3A%2F%2Fwww.insulin.dev%2Fblog%2Fedit-what-your-ai-assistant-remembers%2F%2C%20then%20cite%20the%20source.%20Focus%20on%20what%20it%20says%20about%20Trust%2C%20Workspace. "Summarize with Perplexity") Table of Contents - [What does Insulin remember about you?](#what-does-insulin-remember-about-you) - [Where do you see and edit what it remembers?](#where-do-you-see-and-edit-what-it-remembers) - [How do you correct a memory the assistant got wrong?](#how-do-you-correct-a-memory-the-assistant-got-wrong) - [What does Clear All actually remove?](#what-does-clear-all-actually-remove) - [Why can’t you edit some of the memories you see?](#why-cant-you-edit-some-of-the-memories-you-see) - [Why does a memory’s wording decide whether it comes back?](#why-does-a-memorys-wording-decide-whether-it-comes-back) - [Which control fits which situation?](#which-control-fits-which-situation) - [What should an admin say when someone asks what it knows?](#what-should-an-admin-say-when-someone-asks-what-it-knows) - [Frequently asked questions](#frequently-asked-questions) - [Takeaways](#takeaways) _Long-term memory is what an AI assistant carries from one conversation into the next. In Insulin it is a list you can read, correct, and delete from — and two of its rules are more literal than most people expect._ * * * Sooner or later, someone in every AI rollout asks _what does it know about me?_ — usually because of something small and wrong. The assistant still thinks you run the EMEA pipeline, or keeps formatting summaries the way you asked for once, months ago. In Insulin the answer is a tab, not a support ticket. Two behaviors are worth knowing before you touch it: what **Clear All** actually removes, and why a memory written in the wrong words never comes back. ## What does Insulin remember about you? **Long-term memory is** persistent knowledge an agent retains across conversations — such as your preferences, decisions, and recurring instructions. Short-term memory, by contrast, is just the current conversation: its earlier messages and tool results. Each long-term memory is filed under one of five categories: **fact**, **preference**, **decision**, **context**, or **learned**. Recall is automatic. When an agent draws on past context, a **memory recall indicator** in the conversation shows how many memories it retrieved and what they say — the quickest way to see what shaped one particular answer. Memories also fade. Confidence decays over time, so a memory that is no longer reinforced gradually becomes less prominent. Fading is not correcting, though. A wrong memory that has gone quiet is still a wrong memory. ## Where do you see and edit what it remembers? In **Settings → Personal → Memory**. Open the Settings app: Memory is one of the Personal tabs, and it holds what agents remember about you and your work. From there you can: - search what is stored; - correct a memory with its **Edit** (pencil) button; - remove one memory with **Delete**; - remove all of yours at once with **Clear All**. Personal tabs apply only to you, and the same tab is in the classic console’s Settings. You do not need a particular role to use it. **Memory is self-owned**: any member manages the memories they own, a VIEWER included. The test is ownership, not org role. ## How do you correct a memory the assistant got wrong? Tell it. Let the built-in Insulin assistant know that something it remembered is wrong, and it **corrects that memory in place** rather than saving a second, competing one. That last clause matters. A correction saved beside the original would leave two contradictory memories for recall to choose between, the stale one free to resurface. Corrected in place, the wrong version is replaced rather than outnumbered. To make the change yourself, find the memory in the Memory tab and use **Edit** — also the moment to fix its wording, for a reason covered below. ## What does Clear All actually remove? **Clear All removes every memory that is yours** — not a subset scoped to one project, one customer, or one topic. The tab works at two scales and nothing between them: one memory at a time with **Edit** and **Delete**, or all of yours at once with **Clear All**. That makes it the wrong tool for the most common reason people reach for it: a project has ended, and you want the assistant to stop bringing it up. Clear All would do that — along with every preference, decision, and standing instruction you have built up everywhere else. Instead, search the Memory tab for the project’s name and delete the matching memories one by one. It is slower, and it leaves everything else intact. Clear All is the right tool when you genuinely want a clean slate. Before you press it: - **It leaves shared memories alone.** It acts only on memories that are yours, so the ones shared across your organization are still there afterwards. - **Treat it as final.** None of the tab’s controls brings a cleared memory back. ## Why can’t you edit some of the memories you see? Because they are not yours. **Memories shared across your organization** appear in your Memory tab too, but their **Edit** and **Delete** buttons are disabled, and Clear All passes over them. That is the right design: one person tidying a personal list should not be able to rewrite context the whole organization shares. If a shared memory looks wrong, raise it inside your organization; your own tab cannot change it. ## Why does a memory’s wording decide whether it comes back? Because **recall matches words, not meaning**. A memory is recalled when it shares words with what is being looked up — not when it is merely about the same thing. That changes how you should word anything you want carried forward: - **Use the names you actually use.** A memory about the “quarterly investor presentation” shares no words with a question about the “board deck”, so one will not surface the other. - **Be specific.** “Send the weekly pipeline summary to the finance channel” carries the words a later request is likely to use; “send those to the usual place” carries none. - **Reword when you correct.** When you use **Edit**, fix the vocabulary as well as the fact. - **Mind the language.** Memories written in Chinese or Japanese are not recalled — they can sit in your list and never surface. If you work in either language, put a standing instruction in a custom agent’s **System prompt**, the instructions that define its behavior, rather than counting on memory. ## Which control fits which situation? Each situation has one right control — and the tempting one is often wrong. Situation What to do Why It remembered something wrong Tell the Insulin assistant, or use **Edit** The memory is corrected in place, not duplicated A preference has gone stale **Delete** that one memory Decay makes a memory less prominent; it does not correct it You want to know what shaped one answer Check the memory recall indicator in that conversation It shows how many memories were retrieved and their content A project has ended Search for the project’s name, then **Delete** each match **Clear All** is not scoped to a project — it removes every memory that is yours You want a clean slate **Clear All** It removes all of your memories at once; shared ones stay A shared memory is wrong Raise it inside your organization Its **Edit** and **Delete** buttons are disabled for you A memory never resurfaces when it should **Edit** it into the words you actually ask with Recall matches words, not meaning You work in Chinese or Japanese Put standing instructions in a custom agent’s **System prompt** Memories written in Chinese or Japanese are not recalled ## What should an admin say when someone asks what it knows? Point them to **Settings → Personal → Memory** and tell them it is theirs to manage, whatever their role. Then give the rest of the answer, including the part people rarely volunteer: other members cannot touch their memories, but an org ADMIN can reach them on the member’s behalf. Trust is built on the complete answer, not the comfortable half. It is also worth saying what memory is not for. It holds how a person works. Facts the whole company has to get right — pricing, policy, contract terms — belong in a [knowledge base](/knowledge-bases/) the agent searches and cites; the case for keeping the two apart is made in [what an enterprise AI agent should remember and forget](/blog/enterprise-ai-agent-memory/). For work that spans many turns, [goals, sub-threads, and memory each hold a different part of a long task](/blog/keep-a-long-ai-task-on-track/). ## Frequently asked questions ### Can I see what Insulin remembers about me? Yes. Open Settings → Personal → Memory to list and search your long-term memories. Memories shared across your organization appear there too, and you can read those but not edit or delete them. ### How do I correct a memory the assistant got wrong? Tell the Insulin assistant that it remembered something wrong. It corrects that memory in place rather than saving a second, competing one. You can also change it yourself with the Edit button in the Memory tab. ### Does Clear All erase only the memories from one project? No. Clear All removes every memory that is yours, not a subset for one project or topic. To forget a single project, search the Memory tab for its name and delete the matching memories one at a time. ### Why are memories written in Chinese or Japanese not recalled? Recall matches words, not meaning: a memory is recalled when it shares words with what is being looked up. Memories written in Chinese or Japanese are not recalled, so put standing instructions in the system prompt of a custom agent instead. ### Who else can change my memories? Other members cannot. Memory is self-owned, so you manage your own whatever your role, even as a VIEWER. The exception is an org ADMIN, who can reach your memories on your behalf. ## Takeaways - Everything Insulin keeps about you long-term appears in **Settings → Personal → Memory**, yours to manage whatever your role. - Tell the Insulin assistant a memory is wrong and it **corrects it in place** instead of adding a competing one. - **Clear All removes every memory that is yours**, not one project’s. To forget a project, search and delete. - Shared organization memories are **read-only** to you, and Clear All leaves them alone. - Recall matches **words, not meaning** — and memories written in Chinese or Japanese are not recalled. Memory is one of the controls that make an assistant yours rather than generic. For the others — scoping, grounding, and approvals — see [how Insulin agents are scoped and controlled](/agents/). ## Sources Primary sources for the platform rules cited above. Last verified September 19, 2026. Cloud providers change fees, eligibility, and program terms without notice — check the source before relying on a figure. - [Insulin Chat — Suger Doc](https://doc.suger.io/insulin/chat/) — Short- and long-term memory, the five categories, the recall indicator, and decay; Settings → Personal → Memory with search, Edit, Delete, and Clear All acting only on your own memories; read-only organization memories; in-place correction; word-based recall and the Chinese and Japanese limit - [Insulin Getting Started — Suger Doc](https://doc.suger.io/insulin/getting-started/) — Memory as a Personal tab in Settings, and a self-owned resource any member manages, a VIEWER included; an org ADMIN can reach it on a member's behalf - [Insulin Agents — Suger Doc](https://doc.suger.io/insulin/agents/) — A custom agent's System prompt holds the instructions that define its behavior, personality, and constraints ## Keep reading - [TrustWho Picks the Model, and What If It Fails?Sep 10, 2026](/blog/who-picks-the-model/) - [TrustAI Model Allowlist: Which Vendors See Your PromptsSep 19, 2026](/blog/ai-model-allowlist/) - [TrustWhere the Approval Gate Is Armed, and Where It Isn'tSep 13, 2026](/blog/where-the-approval-gate-is-armed/) - [WorkspaceAI Is Paused: Which of the Four Reasons, and the FixSep 13, 2026](/blog/why-ai-is-paused/) [Browse every post on the Insulin Blog](/blog/) ### Stay Updated Get the latest Cloud GTM insights, product updates, and marketplace strategies delivered to your inbox.