--- title: "Memory Administration and Conversation Exports" slug: "memory-administration-and-conversation-exports" date: "2026-08-24" tag: "Application" summary: "Administrators can now manage every user's memories and every agent's memories from one organization-level tab, including the switches that decide whether anything is captured at all — and every user can review and export their own conversations from their profile." author: "ibl.ai Engineering" linkedin: | Two things are true about AI memory at the same time, and most platforms only build for one of them. A user needs to see what an assistant remembers about them, and take it back. An administrator needs to correct a memory that is wrong, remove one that should never have been kept, and answer the question a regulator or a departing employee will eventually ask. Both shipped on ibl.ai this week. Administrators get an organization-level Memory tab with two views. Global works user by user: open a person and manage the memories that follow them across every agent, along with the two switches that govern capture in the first place — whether the platform may learn from their conversations, and whether what it has learned may shape replies. Agent works agent by agent: open an agent and manage everything it has retained about the people who talk to it, filtered by learner, category, and date. A small detail that matters in a review: memories captured automatically from conversation carry an "auto" badge, so you can always tell what a person entered deliberately from what the system inferred. Users get the other half — a History surface on their own profile. Every conversation they have had with every agent, filtered by agent, date, sentiment, or topic, with the full transcript, a one-click CSV of a single conversation, and a server-side export of any filtered set. It reads endpoints scoped to the signed-in account, so it can only ever show a person their own data. Why we can build it this way: on ibl.ai you own all the code and the data. Memory is rows in your database, inside your perimeter — which is what makes "delete this, and prove it" an operation you can actually perform rather than a support ticket you file with a vendor. #iblai #AgenticAI #EnterpriseAI #DataGovernance #AIPrivacy --- Memory on the platform now has an administrative view, and conversation history now has a user-facing one. The two shipped together because they answer the same question from opposite ends: who can see what the platform has retained, and who can take it back. ## Managing anyone's memories ![The organization's Memory settings on the Global tab, showing a user search above a table of users with Name, Username, and Email columns and a per-row view action](/images/updates/memory-admin-global.webp) The organization's settings gained a **Memory** tab — "Manage user global memories and agent memories" for the whole workspace. It opens on **Global**, a searchable table of the organization's users. ![The Global Memories popup for one user, showing the memory-capture and personalization toggles above that user's saved memories](/images/updates/memory-admin-user.webp) Opening a user shows everything the platform remembers about them across all agents, and — above the list — the two switches that decide whether there is anything to remember at all: - **"Allow AI to learn from our conversations"** governs capture. Off means nothing new is recorded. - **"Use my saved information in responses"** governs use. Off leaves existing memories stored but stops them shaping replies. Turning off capture alone leaves earlier memories in play, which is a distinction worth having: one suspends the effect, the other stops the accumulation. Each memory carries its text, its date, and an edit-and-delete menu. Memories the platform captured automatically from chat show a robot icon and an `auto` badge, so a reviewer can always tell an inference from something a person entered deliberately. Both are equally editable here. ## Managing what an agent has learned ![The Memory settings on the Agent tab, showing an agent filter above a table of agents with Agent and Description columns](/images/updates/memory-admin-agents.webp) The **Agent** tab works the other way round — start from an agent and see what it has retained about the people who talk to it. ![The Agent Memories popup, showing the user filter, date range, category tabs, and memory cards for a single agent](/images/updates/memory-admin-agent.webp) Opening an agent loads the same memory manager its own settings render: a user filter, a date range, category tabs — Knowledge Gaps, Learning Goals, Personal Context, Preferences, and any you add — a category manager, and one card per memory showing when it was recorded and whose it is. There is one memory store behind all of this, reachable from three surfaces with different authority: a user's own profile, an agent's settings, and now the organization tab for someone with the standing to act on another person's behalf. ## Your own conversations, and a way out ![Profile History on the Conversations tab, showing agent, date-range, sentiment, and topic filters above a conversation list and a transcript preview](/images/updates/history-conversations.webp) Every user now has a **History** page on their profile — "Review and export your conversations with AI agents." It gathers every chat with every agent into one filterable place: by agent, by date range, by sentiment, or by topic. ![A selected conversation showing the alternating You and agent messages with a Download button above the transcript](/images/updates/history-transcript.webp) Selecting a conversation opens the full transcript beside the list, formatting and both avatars intact, with a **Download** that saves that one conversation as a CSV built in the browser — no queue, no waiting. ![The Exports tab, listing generated reports with status, creation time, the filters used, expiry, and a download action](/images/updates/history-exports.webp) For anything larger, **Export** generates a report of everything matching the current filters on the server, and it lands under the **Exports** tab with its status, the filters that produced it, and its expiry. Because each row records its own filters, a report downloaded weeks later can still be read for what it actually covers. This page reads endpoints scoped to the signed-in account, so it shows a person their own conversations and nothing else — it is hidden entirely when someone else views their profile. Administrators reviewing a whole agent's conversations use the agent's own Chat History panel instead. ## Documentation - [Organization Settings: Memory](/docs/os/organization-settings/memory) — the Global and Agent views - [Profile: History](/docs/os/profile/history) — conversations, transcripts, and exports - [Profile: Memory](/docs/os/profile/memory) and [Agent Settings: Memory](/docs/os/agent-settings/memory) — the two surfaces this administrative view sits above Both surfaces ship in the ibl.ai SDK, so a product built on the platform can offer its own users the same review and export path.