Find the answer you know you saved
Saving is the easy half. Filters, your own descriptions and a jump back to the source are what make an archive of four hundred entries usable.
For anyone whose problem is no longer saving things, it is finding them again.
The situation
A few months in, an archive has the same failure mode as the chat sidebar it replaced. Four hundred entries, and the one you want is somewhere in there.
Searching a platform's own history does not solve it, because it searches one platform and it searches the model's words. What you remember is what you were trying to do at the time, which is rarely the phrasing that came back.
Model-generated titles make it worse. Half your conversations are called some variation of the same three words, and none of them say what actually came out of it.
What Savelore does about it
- Your own description, searched
- The line you wrote when you saved something is searched alongside the text itself. That is usually the phrase you actually remember, because you are the one who chose it.
- Filters that cut the pile down fast
- Date, model, tags, favourites and pins. Remembering only that it was Claude, sometime in spring, gets you from four hundred entries to a handful.
- Favourites and pins
- The ten entries you reach for most do not deserve to be ranked alongside everything else. Pin them and they stay at the top.
- Folders and tags
- Tag by project or by problem rather than by platform. The platform is already recorded, so tagging it again buys you nothing.
- Chat-aware filtering
- Back in a conversation with the sidebar open, one toggle narrows the list to entries saved from that conversation, and search then only looks inside those. It works in the sidebar on ChatGPT, Claude and Gemini.
- Jump to the source message
- Every entry keeps a link back. Opening it loads the original conversation and scrolls to the message, highlighted, so the context you did not save is still one click away.
- The web dashboard
- web.savelore.app gives the archive more room than a sidebar panel, and makes bulk tag and folder edits easier. It reads the synced copy, so it needs cloud sync switched on.
How to make an entry findable later
Almost all of this is done in the ten seconds after you save, not in the ten minutes you spend hunting six months later.
- 1
Write the description in your words
One line about what this is and why you kept it. Not a summary of the response. The words you would use if you were describing it to someone.
- 2
Tag by project, not by platform
The model and the platform are recorded automatically. Your tags should carry what those cannot: the project, the client, the problem.
- 3
Pin the handful you use weekly
Pins and favourites are for the small set you return to. Using them on everything makes them mean nothing.
- 4
Narrow with filters before you search
Model plus a rough date range usually gets you close enough to just look. Searching a smaller pile works better than searching a big one.
- 5
Use the chat filter when you are back in the thread
Returning to a conversation you worked in before, toggle the filter and see everything you kept from it, then jump to any of those messages.
- 6
Open the dashboard for a clean-up
Retagging and refiling twenty entries is a different job from saving one, and web.savelore.app has the room for it if you have cloud sync on.
The description is the whole trick
It is the one field nothing else can generate for you, and it is the one that works. Everything else about an entry is written by the model or recorded by the software. The description is the only part that holds what you were thinking.
It does not need to be good. One clumsy line about why you kept something beats a well-written summary of what it says, because in six months you will be searching for your own reasoning, not the answer.
Filters beat search when you cannot remember the words
Search needs you to recall a phrase. Filters only need you to recall a circumstance, which is a much lower bar. You usually know roughly when it was and which model you were using, even when the exact wording is gone.
Model, date and tag stack on each other. Two of the three is normally enough to get the list short enough to read.
This is not the same as searching your chat history
Chat history search looks through everything, most of which is noise, on one platform at a time. An archive holds only what you decided was worth keeping, across all nine, with your own words attached.
That difference compounds. The archive gets more useful as it grows because everything in it passed a filter. A chat history gets less useful as it grows for exactly the opposite reason.
The honest limits
Chat-aware filtering works in the sidebar on ChatGPT, Claude and Gemini. On the other supported platforms you have the normal filters and search, but not that toggle.
The web dashboard reads your synced archive, so it shows nothing if you are running local-only. It is also still in beta, and a few export options exist in the extension and not there yet.
Frequently asked questions
What does search actually look at?
The saved text and the description you wrote, so your own phrasing works as a search term even when you cannot remember what the model said.
What can I filter by?
Date, model, tags, favourites and pins, and they combine.
Can I see only the entries from the conversation I am in?
Yes, with chat-aware filtering. Open the sidebar in a conversation on ChatGPT, Claude or Gemini and toggle it on. Search then looks only inside that conversation's entries.
Can I get back to where an entry came from?
Yes. Every entry keeps a source link that opens the original conversation and scrolls to the message it came from. For a collection, each citation links back to its own source message.
Do I need cloud sync to use the web dashboard?
Yes. The dashboard reads the synced copy of your archive, so a local-only archive will not appear there.
Does any of this need Pro?
No. Search, filters, tags, folders, favourites, pins and the source links are all on the free plan. Pro removes the limits on how much you can save.
An archive you cannot search is a folder of files.
The work is not in the saving, it is in being able to get back to it two seasons later. Savelore keeps your words next to the model's, records what platform and model it came from, and holds a link back to the exact message.