Oasis Browser: Long Session Railroad Memory
Long chats break when the model forgets what you said ten turns ago. Oasis uses optional structured session memory (internal nickname “Railroad”) to help follow-ups stay coherent.
Insight for other users: Session memory extracts structured facts across turns and re-injects them into later prompts so the assistant can remember details within your session and plan limits. You do not edit memory by hand. It updates as you chat.
What it is not: Memory is not a full backup of web pages. It is distilled text used to steer replies, not a copy of everything you browsed.
How to work with it:
Keep chatting in the same thread for continuity.
If your build exposes a control, ask the assistant to forget sensitive details you do not want retained.
Start a New chat when you want a clean memory scope (separate from chat history stored in IndexedDB).
Privacy note: Memory sits in the broader assistant data picture. See what leaves the device vs stays local: https://kahana.io/docs/assistant-and-cloud-data
Doc: https://kahana.io/docs/long-session-memory-railroad
Related: https://kahana.io/docs/multi-step-agent-loop | https://kahana.io/docs/chat-history-indexeddb
Questions we would love your take on:
Transparency
Should Oasis show you what it “remembers” in a sidebar, or keep memory invisible?Forget controls
How important is a one-click “forget this” vs starting a new chat?Scope
Should memory last only for the current thread, the whole browser session, or across days?Trust
For a privacy-first browser, does automatic memory extraction feel helpful or creepy?For the Oasis team
In a long thread, did Railroad help coherence, or did you still repeat yourself?
Share how you expect “memory” to behave in an AI browser.
Replies
Long-session memeory is one of those features people only notice when it’s missing 👀 Repeating context every few turns gets exhausting fast.
@malani_willa I think transparency matters a lot here 🔥 Even a lightweight sidebar showing “what the assistant remembers” would build more trust
Oasis Browser
@malani_willa @susie_johns Love these ideas! We can add a light sidebar to show memory capacity :)
It’s interesting that the memory updates automatically instead of being manually curated. That probably improves usability but also increases the importance of good forgetting controls
Transparency wins every time.
Invisible memory feels like surveillance. Visible memory feels like a feature. Even if 90% of users never open the sidebar, the fact that it exists changes how the tool feels to use. Put it in.
On forget controls: one-click forget matters more than people think - not because users will use it constantly, but because knowing it exists removes the anxiety of chatting freely. It's the same reason "unsend" made people more willing to message in the first place.
Scope: thread-only by default, opt-in for session-wide. Cross-day persistence is where you lose privacy-first users fast. Let them graduate into it.
The real trust question isn't "is automatic extraction creepy?" - it's "does the user feel in control?" If they can see it, edit it, and nuke it, most people stop caring that it's there. If it's a black box, even a helpful one, it breeds suspicion.
Railroad sounds like the right call. Just don't hide it.