What your clone can do now that it couldn't before — newest first, in plain words.
Follow along: RSS feed
Your subscription now comes with a clear monthly AI allowance. We email you before you reach it so nothing surprises you, and if you use it all, your clone keeps working, more slowly and within a daily limit, until the 1st, and nothing you send is lost.
Try: nothing to set up. If you get close, that email gives you the reset date, and a big import can be arranged through hello@getklyr.ai.
When the Klyr installer could not be downloaded, the install command used to end quietly and drop you back at the prompt, exactly as if everything had worked. Nothing was installed, and nothing told you.
Now that case can no longer pass for a finished install. The command stops with "Klyr was NOT installed. Nothing was changed on this computer." and tells you what to do next: wait a few minutes and paste the same command again. A finished install still ends with its own closing line, "Klyr is installed and ready" (on Windows, "Klyr installed and ready"), so you can tell the two apart at a glance.
Try: paste the install command from the Getting started page and look at the last line in the window. "Klyr is installed and ready" (or "Klyr installed and ready" on Windows) means you are set; anything else, the Troubleshooting page has an entry for it.
Your app remembers the connection it made to Klyr, even after a reinstall or a sign-in you never finished. If that connection had gone stale in the meantime, signing in again used to stop on a cryptic error page, and the only way out was to clear the app's saved connection by hand.
Now Klyr recognizes the old connection and the sign-in simply continues: you enter your email, click the link, and you are back. In the rare case it cannot, the page tells you in plain words how to give the app a fresh connection, instead of an error.
Try: after reinstalling Klyr, type /mcp, select klyr and choose Authenticate. The sign-in goes through even if your last one was weeks ago.
Connecting your clone to someone else's no longer involves an address you have to look up, copy or share. You add a contact with their email, the one you already use to reach them. There is no other address to look up.
Your contacts see your clone under its own name, the one you gave it. Rename your clone and they see the new name; there is nothing else to update.
In your own list, every contact shows their clone's name. Next to the name you also see the email you typed, for the people you invited yourself. An email is only ever shown to the member who typed it: someone who invited you with your email sees the address they typed, and someone you invited never sees yours.
An invite you send shows in your list with the email you typed, waiting, until they accept (or for 30 days). Changed your mind? Withdraw it with the same email.
Contacts you are already connected to stay connected: nothing you set up is lost. They show by name.
Try: /klyr:contacts and add someone with their email address.
Your Klyr account is your clone: one address, one clone, wherever you sign in.
Now the address decides. Sign in from any machine, or buy again with the same address, and you land in the clone that address already has: same memory, same profile. If you cancel and come back within the 30 days your data is kept, you get that same clone back with its memory, nothing to restore. After those 30 days the data is deleted, and a new subscription starts a new clone.
Ended up with two clones for one address before this change? Write to hello@getklyr.ai and we join them into one.
Try: sign in with /klyr:login on a second machine and ask your clone what it remembers about you; it answers from the same memory as on your first machine.
Klyr for Teams takes its first step. A company can now have a team: every member keeps their own clone and their own memory, and one person, the team admin, decides which of Klyr's four directors (Finance/Admin, Operations, Marketing, Sales) each member's clone works with.
Directors are off by default for a team. Klyr turns them on for the team when it asks; then the admin assigns them per member, or gives none at all. Nothing of your memory is ever part of the team record.
Try: if your company wants a Klyr team, write to hello@getklyr.ai with the address of the person who will administer it. We create the team and name them its first admin.
The Memory screen of the Klyr dashboard shows what your clone learned about you today and which of your notes sit close to each other. On a hosted account that part used to say it was not available yet.
Now it works on a hosted account too, and it only ever shows your own memory.
Try: open your dashboard with /klyr:dashboard, go to Memory, and watch "learned today" fill in once your clone updates.
One Klyr account runs on several machines, and each machine can have several Claude Code sessions open. Until now a message to another of your machines could name the machine, not the session, so it landed wherever that machine decided.
Now you can name the session too. Your message lands in that session when it is open there, and otherwise in the session you used most recently on that machine.
Try: from one session, ask your clone to send a note to your other machine, naming the session that should get it.
On hosted, you can now export what your clone has learned about you: the profile it keeps about you, and every conclusion it ever stored, each with its date. You do it yourself with your own sign-in, and the export always tells you when it has everything. It only ever contains what was concluded about you.
Try: ask your clone "export what you believe about me to a file".
On hosted, every conversation your clone has learned from can now be exported with your own sign-in: each session you took part in, your turns and your clone's, oldest first, with their dates. You no longer need to write to us to get a copy. An export that stopped halfway always says so, and if your history changes while it runs, it tells you to start over instead of quietly leaving something out. It only ever contains your own sessions.
Try: ask your clone "export my conversation history to a file".
On hosted, your memory archive (your notes, your conversation history and every import) can now be exported with your own sign-in. You no longer need to write to us to get a copy. The file uses the same format as a local install, so it is yours to keep or move to a self-hosted Klyr. An export that stopped halfway always says so, and it only ever contains your own entries.
Try: ask your clone "export my Fathom meetings to a file", or "export my whole memory archive to a file".
When a subscription ends, Klyr now emails your account address the same day: when your access stopped, the exact date your hosted memory will be permanently deleted, and how to get your full copy before then (reply to the email). Nothing is deleted unless that email is on record as sent to you, or you were told another way.
The best moment to take your copy is before your paid period ends: until then you can export your memory yourself, with your clone, without asking anyone.
Try: nothing needed. If you ever cancel, watch for the email from Klyr with your date.
If you let the heartbeat check your clone-to-clone mailbox, a new message no longer sits unseen until your next session. The session you are working in gets a one-line notice with who wrote and how many messages. When you are away, the local notification says the same, if you turned it on. A new session also opens with a line saying mail is waiting. Your clone never opens the message by itself: you decide when to read it. A session you started before this update gets the local notification instead, until you restart it.
One small fix around contacts too: if you give your clone someone's email where it needs their address, it now says so and points you to your contacts list, instead of telling you that person is not your contact.
Try: turn on the heartbeat's mail check with /klyr:heartbeat, then keep working in a session when a contact writes to you.
The heartbeat, Klyr's optional background helper, can now run agents you describe yourself: "every weekday at 7, prepare this list", "check that every hour". Your clone writes the agent with you, gives it only the tools the job needs, and the heartbeat runs it on its schedule while you are away. What it finds waits as a short note for your next session. If the agent is edited after you switched it on, it stops until you switch it on again, so it never runs something you did not approve. Each run uses your own Claude usage, and scheduled runs stop while the heartbeat is off. Your HEARTBEAT.md now lists every agent with its schedule, its tools and its last result. Lines you had written in it yourself are kept in HEARTBEAT.saved.md, and your clone offers to turn each one into an agent.
If you chose the local notification, it now tells you where a note came from and what it is about. When there is nothing to name, it stays quiet.
Try: with the heartbeat on (/klyr:heartbeat), tell your clone "every morning at 8, check X and tell me when Y happens". To check the notification still gets through, say "test my notifications again".
If you set up the Klyr window (the launcher that opens your notes folder in its own editor profile), a PDF, a Word document, a spreadsheet or a slide deck in your workspace now opens there as a readable view. Before, the window could only show you that the file was not text. Long file names in the sidebar now scroll sideways instead of being cut. Old .doc and .ppt files are not covered yet.
Try: run /klyr:repair, then open a PDF or a .docx from your workspace in the Klyr window. If it still does not open as a document, ask your clone to set up the Klyr window again.
Your clone keeps the conversation with you and hands the doing (file changes, commands, anything that goes outside) to helper agents. It briefs them with your standing rules, checks what they bring back before reporting it to you, and leaves anything that goes out to other people as your call.
If you already have a clone, /klyr:repair adds this as one new principle to your clone's identity, after a backup, and changes nothing else. If you delete it, it is not added back.
Try: run /klyr:repair, then ask your clone to show you its "Chief of staff" principle. Keep it, reword it, or delete it.
When a Windows PC has no Git, the Klyr installer stops and asks you to run one line in an administrator PowerShell to install it. That line failed, so the install could not go further. It now downloads and installs Git for Windows, with no sign-in needed.
Try: on a Windows PC without Git, run the install command in an ordinary PowerShell, then paste the line it prints into a PowerShell opened with "Run as administrator".
If you asked your clone to keep your workspace tidy, it used to move a note only when its tag or name matched a folder you already had. Most members have no such folders, so most notes just waited. Now your clone reads each unfiled note and files it: into a folder you already have, or into a new, clearly named one. Every run is recorded, and one request undoes it, removing any folder it created once it is empty. A filing you undid is never repeated.
This filing runs with the heartbeat, Klyr's optional background helper. It now also works on Windows and Linux, and /klyr:status says plainly whether filing is ready.
Try: turn on tidy filing with /klyr:config and the heartbeat with /klyr:heartbeat. To reverse a run, say "undo the last filing run".
Before asking you for a fact, such as who, when, how much or where, your clone now looks first: your memory, your meetings, your mail and calendar, connected business systems and the web. If it still asks, it tells you what it searched and what it found.
Questions that are really decisions ("do you want me to…", "shall I…") still reach you as asked. This is on by default, and you can turn it off.
Try: ask your clone something whose answer is in your own notes. To turn this off, tell /klyr:config "let it just ask me".
Setup used to end by handing you a folder path. On macOS and Windows, your clone can now give you a way in: a Klyr launcher that opens your notes folder in your code editor, in its own Klyr profile set up for reading and writing notes. Your other folders and your usual editor setup are left as they are. Klyr does not install an editor; it uses the one you have.
The Klyr window opens light, and your clone's own theme, Klyr Light, matches it inside Claude Code there. Anything Klyr set up can be removed again, and settings you chose yourself are never overwritten.
Try: ask your clone "set up the Klyr window". If the text looks washed out or you prefer dark, ask it to switch the Klyr window's theme.
When you ask your clone to recall something, sessions your background agents ran now come after your own notes. They are still searched and can still come up, but an agent's transcript that happens to use the same words no longer pushes a note you wrote yourself off the list.
A recall that comes up empty is also no longer read as proof you never wrote it down. Your clone now knows a recent note may simply not be searchable yet, and says so instead of telling you there is no record.
Try: ask your clone about something you noted today. If it finds nothing, it now tells you the note may not be searchable yet.
You can keep your clone's memory on a Windows PC, the same way as on a Mac or a Linux machine. Local memory needs Docker Desktop, left on its default settings.
Klyr's machine check on Windows also stopped telling you your graphics card was too weak. It now asks you whether your machine is powerful enough instead of guessing, so you can carry on with Local.
Local memory still needs a strong machine, on every platform. Hosted memory needs none of this.
Try: run /klyr:init and choose Local. On Windows the machine check now asks you instead of blaming your graphics card.
If your memory service had a problem, your clone could answer as if your memory were simply empty. That looked exactly like losing everything. Now an outage is reported as an outage: your clone tells you it could not reach your memory, says plainly that this is not an empty memory, names which part failed, and points you to /klyr:status.
A search where one source failed and the others found nothing no longer comes back as "nothing found" either.
Try: nothing to do. If your clone ever says it could not reach your memory, run /klyr:status to see which part is down.
Klyr updates itself in the background, so you do not need a command to receive changes. What you sometimes need is a fix when something has drifted. That command is now called /klyr:repair: it checks your install, repairs what has drifted and cleans up old files. /klyr:update still works and does exactly the same thing.
Running the install command again now also gets you the latest release on your track. Before, it could hand back the older release already on your machine, the very one that was not starting. An older install that still kept its settings in a previous location is also moved over automatically, so your settings and onboarding answers come with you.
Try: if something looks wrong in /klyr:status, run /klyr:repair.
On hosted Klyr, your clone can now put a question to a contact's clone and bring back the answer. The answer is written on their side, from their memory, and only the answer crosses: no documents, no files, no list of sources.
It only works if they said yes. Being connected is not enough: each member decides, contact by contact, whose clone may question theirs, and can take that back at any time without disconnecting. If you ask someone who has not said yes, you get a polite refusal.
Try: /klyr:contacts, then "let
Connecting two clones used to start with a chore: before anyone could add you, you had to pick a name for yourself and hope nobody had taken it. Now your clone is given its own address the moment it exists, so you are reachable straight away, and nobody loses a good name to whoever asked first.
The name people actually read is still yours. Set your display name to whatever you like, change it as often as you like, and your contacts see that — the address underneath simply does the routing. If you already picked an address, you keep it exactly as it is; nothing you chose was renamed.
Try: /klyr:contacts and look at the top line: that is the name your contacts see.
When you moved, renamed or deleted a Google Doc, the copy your clone keeps could stay behind at its old place. Your clone could then find an old version and present it as current. The Drive sync now follows those changes: a moved Doc moves, a renamed one is renamed, and a deleted one leaves your clone's copy too.
Each copied note now links back to its original Doc.
Try: rename a Doc in a folder you sync, wait for the next sync, and ask your clone about it by its new name.
On Windows, Klyr's background helper used to run in a terminal window you could see. Closing that window stopped it, and your clone lost its memory until the next login. It now runs with no window at all. A machine that still had one open fixes itself the next time the helper starts.
If the helper stops for any reason, it now restarts on its own, instead of leaving you without memory until you log in again.
The Windows install also got sturdier. It now really installs the background helper (some installs stopped before that step). And when something goes wrong, it shows you the error instead of closing your terminal.
Try: nothing to do. If you ever closed a Klyr window on Windows, run /klyr:status to confirm memory is connected.
Two things the get-started page was getting wrong, depending on what you run.
On Linux, the command it gave you could stop on its first line without installing anything. It now installs.
On Windows, it offered self-hosted memory as a free choice without saying what it needs: self-hosting there requires Docker Desktop. The page says so now, where you make the choice, instead of after.
Try: if an install ever stopped on you, run the command on /docs/getting-started again.
A new clone no longer starts from blank headings. It comes with working rules: say what it knows and what it does not, ask only when the answer changes the work, report what actually happened, and confirm before anything that leaves your machine or is hard to undo. You do not need to be good at writing prompts to get a clone that behaves well.
If your clone was set up before, nothing in your file changed. Your clone can show you each of these rules and add only the ones you pick, at the end of the right section, with a backup first. It never rewrites a line you wrote.
Your clone also reminds you when it has been a long time since it last interviewed you (every 90 days by default), because what it knows about you goes stale. You can change that interval or turn the reminder off.
Try: ask your clone "show me the default rules you can add to my profile". To change the reminder, ask your clone through /klyr:config.
Your clone ships with two helpers you can call on any time. The devil's advocate takes a plan, a decision or a claim and makes the strongest case against it, then tells you what would have to be true for it to fail. The cold reviewer reads something you are about to send exactly as the person receiving it will, with no knowledge of the conversation behind it, and tells you what they would misread or push back on.
Both only read and report. They never decide for you and never rewrite your work. And both are allowed to say nothing is wrong, so when they do raise something, it is worth reading.
Try: before you send your next proposal, tell your clone "read this cold". Before your next big decision, say "play devil's advocate".
Installing hosted Klyr used to need two sign-ins: one for the install, then a second one inside Claude Code before your clone could reach your memory. Now the install's single browser sign-in covers both. Your clone connects to your memory straight away, with no extra step.
Running the install again on a machine that is already signed in no longer sends you back to the browser either. It checks that your sign-in still works and carries on.
Try: run the install command from /docs/getting-started again. If you are already signed in, it skips the browser.
On hosted memory, your clone now reads what you type before it answers and pulls in the notes and past sessions that match it. Before, it started each turn with the same general context, whatever you asked. Now a question about a client or a project brings back what you wrote about that client or project, in the same turn.
Try: ask your clone about something you wrote down weeks ago, by name, and see it answer from your own notes.
When your clone recalls a note, it now tells you whether it's still a draft or something you've approved — so an unverified thought never travels as settled fact. Your clone leans on what you've confirmed and flags the rest, keeping your memory honest by default.
Try: ask your clone about something recent and notice it marks unconfirmed notes as drafts.
Pick the release track that fits you: stable (vetted, the default), beta (early access once a release has baked), or canary (bleeding edge, for builders who want every change first). Switch anytime — your memory and your clone stay exactly as they are, a track change only decides which updates you receive.
Try: run /klyr:channel to see your current track, or /klyr:channel beta to move up.
The background helper that runs your clone's memory now updates itself. You set it up once, and after that it keeps pace on its own — no command to run, no version to chase. Improvements reach you quietly, without an update dance.
Try: nothing to do — it's automatic.
Your clone now picks up what you discuss as you go, not only when you close the session. Bring up something from a chat a few minutes ago — even from a different session — and it's already there to recall, instead of waiting until you're done. Your recent context stops living in a session you haven't closed yet.
Try: in a new session, ask your clone about something you talked through a few minutes earlier elsewhere.
Your clone's memory is now reachable from any Claude surface — including a browser where Klyr can't see a single file on your machine. Add Klyr as a connector once, and from your phone, Claude Desktop, or a cloud code session you can ask what you've built together and it recalls. Nothing to install on that surface.
Try: open Claude in your browser, connect Klyr, and ask it what it remembers about a recent topic.
Introduce yourself to a fellow member and your two clones exchange context — with both of you in control of what's shared. No group thread to manage, no copy-paste of who-knows-what. Your clone brings you up to speed on them, theirs does the same for you.
Try: ask your clone to invite a fellow member, then let them accept from their side.
Imported the wrong thing, or something landed under the wrong label? Ask your clone to fix it. It counts what would change first, shows you, and only deletes once you confirm. On hosted memory, deletes work by project or by session for now; deleting a whole import at once works on local memory only. Your memory stays yours to curate — no support ticket, no waiting.
Try: ask "what did that last import bring in?" and clean up anything that doesn't belong.
Point Klyr at the Drive folders that matter and it keeps them in sync on your schedule — every 15 minutes, hourly, or daily, your call. Read-only, nothing streams anywhere it shouldn't, and a first big sync runs in the background instead of blocking you. You keep working; your clone stays current.
Try: run the connector setup, pick a folder or two, and choose a cadence.
If you ever cancel, you get a 30-day window from the end of your paid period to get a full export of your memory before it's permanently deleted. During that window the export is sent on request; while your subscription runs, you can export it yourself. No lock-in, no games — your data is yours to take with you.
Try: nothing needed today. If you cancel, we email you the date your window closes and how to get your copy.
Ask Klyr how it's doing and you get a straight answer — what's working, and what isn't — instead of a green light that hides a problem. If something needs your attention, it says so plainly, and tells you the one thing to do about it.
Try: run a status check and see the honest picture.
Your memory now shows its sources — which import, which conversation, which document a piece of context came from — in language you can read, not internal codes. When something looks off, you can see why it's there and decide what to keep.
Try: ask your clone where it learned a particular fact about you or your work.