MementoWhat’s new
Changes to Memento worth knowing about, newest first. Not every release appears here — only the ones that changed something you can see.
Network requests no longer cover your screen
Fixed · · v0.10.109
When the assistant wants to reach a website you have not approved yet, Memento asks you to allow or block it. Every waiting request used to get its own card, stacked in one column, and with several waiting on a phone they covered the bar along the bottom, so you could not go anywhere until you had answered them all.
Now you see one request at a time, with arrows to step through the rest and a count of how many are waiting. Minimize puts them away into a small button with the count; nothing is allowed or blocked until you decide, and requests that arrive in the meantime light the button up instead of opening over what you are doing. On a phone the card sits above the bottom bar instead of on top of it.
Out-of-date changes and failed answers say so honestly
Fixed · · v0.10.108
A change waiting for your approval can go out of date when Memento rewrites the same lines of a page in the meantime. Approving one used to fail with a bare “internal error”. Now the card in your inbox, the review and the home page mark it out of date before you click, name the pages that changed, and offer Re-run, which files the capture again against your current library, or Discard.
When the assistant could not answer — your subscription limit was reached, the model was refused, or the run timed out — chat could show the error text as if it were the answer, and keep it in the conversation after a reload. A failed turn is now shown as failed, is not saved, and after a reload the question simply reads as unanswered. The message also says what went wrong, the same way in chat, inline AI and filing, instead of always asking you to reconnect.
Inline AI no longer touches your selected text when a rewrite fails. Before, a failure message could briefly replace your selection, long enough to be saved into the page. A Claude rewrite now appears all at once when it is finished, instead of word by word.
The assistant can use the connectors you added at claude.ai
New · · v0.10.106
If Claude is signed in with your claude.ai account, the assistant now has the connectors you already added there — Gmail, Google Calendar, Google Drive, Notion and the rest — in every Claude space, with nothing to set up in Memento. Before, Memento hid them from the assistant.
They start read-only, like every connection: the assistant can search and read, but cannot send or delete mail, change events or edit documents. If you want it to, tick Allow write tools under Your claude.ai connectors in Settings → Connections. An API key or a setup token does not carry claude.ai connectors; sign in with your claude.ai account under Settings → Coding agents to get them.
Fixed: when your Claude sign-in had been revoked, chat could show Claude’s login error as if it were the answer, and keep it in the conversation. It now says the sign-in needs renewing and stores nothing.
New models show up in the model picker on their own
Improved · · v0.10.105
The model picker in Settings now lists what your own account can use, read from Anthropic or OpenAI, instead of a list we kept by hand. New Claude models appear within a day, including for Claude Pro and Max subscriptions, which used to wait for an update from us. Codex models appear after your next Codex run, without signing in again.
Inline AI in the editor can now use the GPT-6 models — GPT-6 Sol, GPT-6 Luna and GPT-6.1 Sol. Before, a space set to one of them got an error from inline AI, even though chat and filing worked. Inline AI with Claude Haiku 5.5 also reports its context size and cost correctly now.
A home that shows what needs you and what was filed for you
New · · v0.10.103
Opening Memento now lands on a home page instead of an empty chat. On the left, Needs you lists what is waiting on you: changes to approve, captures with a question, filings that stopped, and drafts, each with its buttons right there. A change waiting for approval shows the sentences it would add, so you can approve it without opening the full comparison.
On the right, Filed for you shows the pages Memento wrote since your last visit, or this week on your first. Each one quotes what was written, with the sentences that were added underlined, and links to the source they came from.
The ask box is still at the top. Your earlier conversations are where they were, in the conversation menu. Page names and quoted page text are now set in a serif, so your library reads like a library again.
A bell for your notifications, and notifications that open what they are about
New · · v0.10.102
There is now a bell at the top of the Library — Alerts in the bar along the bottom on a phone — with a count of the notifications you have not seen. It opens a list of your recent notifications, newest first, read and unread, each with when it arrived. Mark them read one at a time, or all at once.
A notification now takes you to what it is about: a routine’s result opens its conversation, a capture that needs an answer opens its card in your inbox, a change waiting for approval opens the review, a reminder opens the page with the task, and suggested rules open Settings → Instructions. Opening one marks it read.
The pop-up that announces a new notification now stays up for 15 seconds, and for as long as your pointer is on it, with an Open button that does the same. Before, it disappeared after four seconds with nothing to click, and the only list was at the bottom of Activity, showing unread notifications only — once read, a notification was gone for good. Notifications from before this update are still in the list; they just have nothing to open.
Connect Notion, Linear, Jira and more in one click
New · · v0.10.101
Settings → Connections now has a catalog of services the assistant can look things up in — Notion, Jira & Confluence, Linear, monday.com, ClickUp, Todoist, Granola, Fireflies, Otter, Miro, Airtable and Sentry. Click Connect, sign in on the service’s own page, and you are back in Memento with it switched on for your space. Before you go, Memento shows what the agent will be able to do and which address its sandbox will be allowed to reach.
Every connection starts read-only: the agent gets the tools that search and read, and nothing that can change things in the service. Writing is a separate switch per connection, because changes made in another service do not pass through Memento’s review the way edits to your library do.
Any other MCP server can be connected by its address alone — Memento finds out how it signs people in. When a service revokes or expires a sign-in, the connection now says Needs reconnect instead of Connected, runs carry on without it and tell you so, and disconnecting revokes the sign-in at the service where it allows that.
Thumbs up and down, and rules Memento learns from your corrections
New · · v0.10.100
Answers in the assistant, changes in Activity and the analysis of each capture in your inbox now have a thumbs up and a thumbs down. Add a note if you like — it is private to you.
When you correct Memento — a thumbs-down with a note, a change you reject or undo, a category you change on a capture — it looks for a pattern, at most once a day per space, and may suggest a short standing rule such as “File receipts under Finance, not Topics.” Suggestions appear in Settings → Instructions. Nothing changes until you add one, and you can reword it first.
An added rule becomes part of that space’s instructions, so filing, chat and routines follow it from the next run on. A rule that would break the page format or your task list, or contradict Memento’s own safety rules, is refused with the reason.
Also fixed: when Memento could not make sense of its own analysis of a capture, it used to file it anyway. It now waits in your inbox for you to check.
Routines that only tell you when something changed
New · · v0.10.99
Each routine now has a Notify me setting in Settings → Routines: every time, only when something changed, or never. “Only when something changed” is for routines that watch rather than report — “tell me if the status of X changes”, “any new papers on Y?”. A run that finds nothing new stays quiet; a run that finds something sends one notification with a single line saying what changed.
Quiet runs still happen. Each one adds its answer to the routine’s conversation with a short note saying whether anything was new, and Settings shows every routine’s latest verdict. If the assistant cannot say clearly whether anything changed, you are notified anyway, and a failed run always tells you unless the routine is set to never.
Until now every run sent a notification, including all the mornings the answer was “nothing new”. Your existing routines keep notifying every time until you change them, and their notifications now carry a one-line summary instead of the first few sentences of the answer.
Capture by email: forward anything to your own Memento address
New · · v0.10.98
You now have a personal email address in Settings → Email capture. Forward a newsletter, a receipt, a thread or a message with a PDF attached to it, and it lands in your inbox as a capture — the same as sharing it from your phone. The subject becomes the title, the sender and date are kept, and attachments Memento can read come along in the same capture.
You choose which space mail goes to, and whether it is analysed and filed or simply kept as reference. Email has a daily limit of its own so a mailing list cannot use up your agent runs; past it, mail is returned to its sender and you get one notification.
The address works like a password. If it ever leaks, regenerate it and the old one stops working at once.
Choose which model Codex uses for a space
New · · v0.10.91
Spaces that run on Codex now have a Model setting in Settings → Agent, as Claude Code and opencode spaces already did. The list is the models OpenAI offers your own ChatGPT account, and “Default” names the model the space actually uses today.
Nothing changes on its own: every space keeps the model it has been using until you pick another. Until now Codex settled on a model once, when you connected it, and there was no way to see which one it was or to change it short of disconnecting.
Change a category’s icon after you have made it
Improved · · v0.10.89
In Settings → Categories, the icon beside each category is now the icon picker itself. Click it, pick another, and the category — and its section in the Library — changes straight away. This works for the built-in categories too, so People, Meetings or Inbox can look like whatever you think they are.
Until now the icon was only choosable in the moment you created a category, and fixed forever after. The only way to change one was to delete the category and make it again, which leaves the pages already filed there behind, under “Other”.
Memento now tells you when a connected AI needs reconnecting
Fixed · · v0.10.88
When the AI you connected stops accepting your login — you signed out elsewhere, or the subscription changed — Settings now shows that connection as needing a reconnect, with a button to fix it. It used to say "Connected" no matter what, and the only way to find out was to watch your work fail.
That gap was worst for ChatGPT, which is what new spaces use by default: its connection had no health check anywhere in Memento, so a login that had stopped working looked perfectly fine and stayed that way.
The first screen you see when setting up had the same fault, and one worse consequence. It greeted a lapsed login with "Connected and ready.", then sent you straight into your library instead of asking you to sign in again — so you arrived somewhere that could not run anything, having been told everything was fine. It now says the login expired, and takes you to the sign-in step when you pick it.
Claude Fable 5.1 can now be chosen for a space
New · · v0.10.87
Anthropic released Fable 5.1, their most capable model, and you can now pick it for any space in Settings alongside Opus 5 and Sonnet 5. Nothing changes on its own: every space keeps the model it already uses until you change it.
Memento runs the model on your own Claude credential, so which models you are offered depends on what your key or plan can reach. If yours cannot use Fable 5.1, a run with it selected stops with a message saying so, and it is dropped from your list rather than offered again.
A busy assistant now says so, instead of failing with an error
Fixed · · v0.10.84
Memento does one thing at a time on your behalf, so that two runs can never edit your pages at once. If you ask for an edit while it is still filing something you captured earlier, it now waits a moment and then tells you plainly that your assistant is busy and to try again shortly.
Before, the same situation ended the edit with an internal message that began "lock timeout for user" and quoted an account number. Nothing was wrong with your request or your notes — the work you had already set going simply had not finished yet, and asking again a minute later always worked. Nothing on screen told you that.
Saving the same document to a second space no longer does nothing
Fixed · · v0.10.83
Saving a document you had already saved somewhere else now files it in the space you picked. Before this it quietly did nothing: the confirmation screen said it was captured and named the space you chose, and then nothing ever arrived there.
Memento keeps one copy of a document per space, so that saving the same thing twice does not split your notes in two. That check was right, but the identity behind it ignored which space you were saving into — so a perfectly good second copy was mistaken for a duplicate you already had and dropped, while the screen still reported success.
The sample space new accounts start with was reached by the same fault and could arrive without its example captures. It no longer can.
Answers say when they lean on something you saved
Improved · · v0.10.81
When Memento answers a question using a page you added as reference — an article, a spec, a paper — the link it cites now carries a small mark, and hovering it tells you where the material came from.
Answers drawn from your own pages look the same as always. The difference matters: a page you wrote or had filed has been read and distilled, while something you saved to read later has not been checked by anyone yet. Both are fair to answer from, but you should be able to tell which one you are looking at.
Reference material has its own place
Improved · · v0.10.80
Anything you add as reference — an article, a manual, a paper you want to keep and search but not have rewritten — now has its own Reference section in the library, instead of being mixed in with Sources.
Sources holds two unrelated things: the untouched originals of documents Memento has already read and filed, and the material you saved on purpose to read yourself. Nothing distinguished them, so the things you had deliberately put there were the hardest to find again. Each reference now shows the site it came from, and opening one says plainly that it has been kept as written and never rewritten into your pages.
References you added before this change were moved across for you.
The sample space no longer uses up your free one
Fixed · · v0.10.79
The sample space Memento builds for you on the first day was counting against the free plan’s limit, so the demo sat in the slot you wanted for your own library.
It is a grant now. You get it in addition to your own space, not instead of it.
The free plan runs twenty agent tasks a day
Improved · · v0.10.78
It was five, which turned out to be too few to actually try Memento — you could capture a handful of things and hit the wall before the library had taken any shape.
Inline AI editing works whichever agent you connected
Fixed · · v0.10.75
Asking for an inline edit inside a page failed for anyone running Claude Code or opencode. Only Codex spaces worked.
The editing feature runs the agent CLI directly, and two of the three CLIs were missing from the server image — so the request died before it reached your agent. All three are now present and checked at build time, so this cannot come back quietly.
Report a bug without leaving the app
New · · v0.10.73
There is a "Report a bug" item in your account menu, and if the app ever crashes on you, the error screen offers the same box while you still remember what you were doing.
A report carries your account email, the page you were on, your browser and the app version. It does not carry a screenshot, your notes, or anything from your library.
Send a share straight to your reference shelf
Improved · · v0.10.70
The space prompt you get when sharing into Memento now offers the reference lane too — for material you want kept and searchable, rather than read and rewritten into your own pages.
Capture from your iPhone’s share sheet
New · · v0.10.69
Add the Memento Shortcut once — “Add to iPhone” in Capture now hands you the finished one — and anything you can share on iOS goes straight into a space of your choosing without opening the app: a link, a photo, a PDF, selected text.
Shared links are no longer cut in half
Fixed · · v0.10.68
A link shared into Memento was truncated at its first "&", so anything with more than one query parameter — most search results, most video timestamps, most tracked URLs — arrived broken and the agent read the wrong page, or no page at all.
Links now arrive whole. One limit is worth knowing and cannot be fixed from our end: anything after a “#” never leaves your phone, so a link to a named section captures the page it sits on.
The app moved to the top of its address
Improved · · v0.10.62
Memento is now at app.mementokeep.com, rather than a /app path underneath it. Shorter, and it matches where the app actually lives.
Old bookmarks pointing at the /app path will not find anything — please re-bookmark from the address bar.
Emptying a page now asks before it sticks
Fixed · · v0.10.60
Select everything in a page, delete it, and the empty version was saved over your text about a second later — by ordinary autosave, while you were still sitting there. Most people found out on the next visit, when the page came back with nothing in it.
A save that blanks a page which still had content is now refused outright, and the editor asks instead: empty it if you meant to, or put back the version still held for you. Emptying a page is still allowed, because sometimes people mean it. It just cannot happen quietly any more.