@withbuddi/buddi 0.1.0-pre.41 → 0.1.0-pre.43

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@withbuddi/buddi",
-  "version": "0.1.0-pre.41",
+  "version": "0.1.0-pre.43",
   "type": "module",
   "description": "Your personal agents, on your computer",
   "repository": {
@@ -17,7 +17,7 @@
     "buddi": "packages/install/dist/launcher.js"
   },
   "buddi": {
-    "notes": "- First run: adding a mailbox is now a short sheet instead of the whole Mail settings page. Pick Gmail, iCloud, Fastmail or Other, type the address and app password (with a link to where to make one), and it checks the login and closes once Mail Triage is reading it. If you took on My days, chapter 4 also asks for your calendar's private link (or a Google sign-in) the same way; if you took on My money, it says plainly that Finance reads the statements you hand it rather than signing in to your bank.\n- First run, Telegram: pairing your phone opens a sheet with a large QR code on a white tile you can scan from across the desk, your bot's name under it, Open in Telegram and Copy link (no bare link to squint at). \"Waiting for your Start…\" turns into \"Paired with Amen's phone\" as soon as you press Start, and the sheet closes by itself.\n- First run inside buddi.app: the \"app and browser\" card no longer tells you to add buddi to the Dock from Safari. It says buddi is already in your Dock, points to Start at Login in the buddi menu, and links the Chrome extension.\n- Signing in with ChatGPT: the code appears once, large, in its field with Copy, and the sentence reads in the order you do it (\"Open openai.com, type this code there and approve buddi\"). No tab opens before you've seen the code; after Open it says \"Waiting for your approval…\", then \"Signed in as <your address>\". Settings → Model accounts and the GitHub connection sheet read the same way.\n- Chat: New conversation, opening a conversation from the list and first run's Open buddi put the cursor in the message box, so you can just type (not while an approval or question card is showing). In buddi.app the window gives the page the keyboard whenever it comes to the front.\n- First run, \"Who do you want on your team?\": each suggestion says why it's there (\"because you set up My days\"; the others are marked popular). A suggestion that isn't ready yet stays, greyed, with its one missing step (\"Chief of Staff · ready once you add a mailbox\") and a button for that step, instead of quietly disappearing; and the card no longer says the picks came from your setup when only popular ones are showing.\n- buddi.app updates: a new buddi is offered in one place only, the app's own updater — one quiet \"buddi X is available\" alert per version and \"Update to X\" in the menu bar; Settings → System shows the same version and its button does the same thing. The Sparkle app updater now only offers a new app (when the app itself changed), asks before downloading, and Check for Updates… no longer shows two windows.\n- buddi.app: when buddi from npm is still running in the background, the window's \"Another buddi is already running\" screen has Take Over right there, and says the npm copy keeps running buddi meanwhile. Choosing Not Now no longer hides the offer for good: the app asks again the next time it opens.\n- When another program has taken buddi's port by the time buddi starts again, buddi moves to a free one instead of failing, remembers it, and says so in one line (\"Port 4317 was taken by another program; buddi now listens on 4391\") in the app's menu and in Terminal; Settings → Where agents may look offers to pair your Chrome extension again with the new address. A port held by buddi's own dashboard, left running after a crash, is taken back instead of moved.\n- buddi.app: buddi → Install Command Line Tool… (or the button in Settings → System) puts a `buddi` command in /usr/local/bin (macOS asks for your password once; otherwise ~/.local/bin, with how to add it to PATH). It runs the app's own buddi and keeps working after updates. It won't install while buddi from npm is on your PATH and says where that one is; `buddi doctor` says which buddi answered and warns when the app a command points at is gone. buddi from npm, run while buddi.app is running, no longer sets up a second background service (\"buddi is already running from buddi.app; this command line talks to it\"), and its `buddi upgrade` points you to the app instead of updating. With buddi.app quit, the `buddi` command opens the app rather than starting a background service of its own. Install Command Line Tool also recognises npm's buddi behind Volta or asdf, and warns when another `buddi` comes first on your PATH.\n- Uninstall from the app: Settings → System ends with \"Remove buddi from this Mac\", and buddi.app's menu has buddi → Uninstall buddi…. Both list what goes, take one last backup into ~/buddi-backups, show its passphrase with Copy and an \"I wrote it down\" box before anything is deleted, and offer \"Keep my data for a reinstall\"; then buddi stops, is removed, and buddi.app moves itself to the Trash and quits. It only works from the computer buddi runs on. If the removal fails, buddi.app stays, says why and offers Retry; an error from Remove buddi shows on the passphrase step (instead of \"Removing buddi…\" for ever) and the confirmation stays good for another try. The six words are kept for the page only until the removal starts or 30 minutes pass, and a Keep my data removal is never repeated by a later restart.\n- Backups: the six words that open your backups are no longer something to deal with during first run. After the first backup (scheduled or by hand), Home shows them once in a card — \"Write these down; they open your backups and only you have them\" — with Copy and \"I saved it\", and it stays until you press it; a paired phone gets them once on Telegram, with a reminder to delete the message. After that, Settings → Backup shows them with Reveal, behind your PIN when you have one (the Home card is shown only on the computer buddi runs on, and asks for your PIN first when you have one), and restoring a locked backup can read the `.passphrase.txt` that uninstall leaves beside it.\n- Uninstall: buddi's command line, run straight from inside buddi.app (or from a copy that isn't a source checkout), no longer mistakes itself for a developer checkout. `buddi uninstall` and `buddi service install|stop|uninstall` refuse with one line (\"Run this from the app's menu, or set BUDDI_DATA_DIR to the installation you mean.\") instead of removing another installation's keychain entries, and uninstall names the folder and keychain it is about to empty and checks them against the installation's own record first. `npx @withbuddi/buddi` run inside someone else's pnpm monorepo is not taken for a checkout either, and a buddi folder moved or restored to a new place is recognised at its next start.\n- Uninstall: before your secrets are removed, `buddi uninstall` prints your backup passphrase and saves it next to the last backup in ~/buddi-backups as `<backup>.passphrase.txt` (only you can read it). `--copy-to ~/Desktop` puts a second copy of the backup and the passphrase there; `--no-backup` still prints and saves the passphrase unless you add `--i-have-the-passphrase`. `buddi vault delete BACKUP_PASSPHRASE` does the same, and restoring a backup uses the `.passphrase.txt` beside it. If the passphrase can't be read (a locked keychain), uninstall stops with nothing removed unless you add `--i-have-the-passphrase`.\n- First run: \"What should I take on for you?\" only offers what buddi can actually install today. My code is gone until the Developer plugin is published, so ticking it no longer ends in a failed install; if withbuddi.com can't be reached, you're offered the plugins known to be published.\n- Replies from Ollama models no longer show the model's thinking before the answer; it is kept apart, shown folded like Claude's. With Thinking off, gpt-oss now thinks as briefly as it can (it has no off switch). A one-word reply such as \"An\" is no longer swallowed."
+    "notes": "- Chat: dragging the edge between the conversation and the Canvas is smooth now, works with a finger, and settles gently at a third, a half and two thirds. A thin blue bar shows the edge when you point at it. Double-click it to go back to the usual layout; with it focused, ← and → move it. Neither side gets too narrow to use, the conversation is drawn exactly as wide as you set it (it no longer comes out a little narrower near the edge), and when the window narrows it gives way so the Canvas keeps its room, then comes back. The width you chose is kept on this device.\n\n- Chat, the message box: Thinking is a small spark button with its dot (hover says \"Thinking: on\" or \"off\"), so the model's name keeps its place. The model chip no longer disappears in a narrow chat; it cuts its name short, and in a very narrow box shows just the short name (\"sonnet-5\"). Send is never cut off.\n\n- Chat: the list of your team hides from the chevron at the right of its \"Your team\" line (or ⌘\\ — Ctrl+\\ off a Mac). Hidden, it leaves a thin tab on the chat's left edge that brings it back, with a dot when someone is waiting for you. The empty row under Agent Father is gone.\n\n- Canvas: a tool's result without a view of its own is a tidy card now — its name in words you'd use, with the agent's face; facts in a clear grid; yes/no and states as small badges; short lists as chips; nested details folded; dates and times your way. A profile leads with the name, a list shows each item with a fact or two, and settings read as rows. Raw JSON is under the card's ⋯ menu. What an agent reads to know you (your profile, the first-run state, memory lookups, your team) no longer opens a Canvas tab by itself; open its step in the conversation to see it. What it changes (your profile, your name, a note, a preference, finishing first run) still gets its tab.\n- First run's handover suggests teammates from what you actually picked in chapter 3 (Pictures → Illustrator, My money → CFO, My days or My mail → Chief of Staff), even if the page was reloaded in between, and says who is already on your team.\n- A plugin with a teammate in the catalogue that nobody on your team uses now says so where you are: its Settings page opens with \"Nobody makes pictures yet · Add Illustrator\", and Home suggests that teammate (with a × for not now) until you add it. Closed, it stays closed, on a new team too.\n- First run: adding a mailbox brings in Mail Triage on the same press (the sheet says so only when Mail Triage isn't on your team yet), so \"Mail Triage is reading it\" is true when it says so, and the handover names it among your team. Home's \"Background triage needs a mail agent\" no longer shows up during first run; it appears only afterwards, when you have a mailbox and no mail agent. Nothing waits for your approval on a clean first run.\n- First run, chapter 4: the calendar row shows the Google account you signed in with (\"Google · you@gmail.com\"), \"Waiting for Google…\" while a sign-in is under way, and asks you to sign in again when Google stopped accepting buddi's sign-in.\n- First run: the plugins chapter 3 installs in the background (Calendar, Weather) show up in the rail and in Settings as soon as they are in, without a reload or a restart: buddi tells every open window when a plugin is loaded, enabled or disabled, and the handover's Open buddi reads the pages again. If one really needs a restart, the handover and Home say \"Calendar is installed; it wakes up after a restart\" with a Restart button. `buddi doctor` lists the pages the running buddi has next to what is installed, and warns about one that is installed but not loaded; it no longer trusts a leftover record from an earlier run whose process number was reused. Once first run is over and nothing waited for a restart, Home stops asking. Reload is in the owner menu inside buddi.app too.\n- Settings → Model accounts: Test connection now asks the account's default model to reply with the single word \"ready\" and tells you in one sentence what happened (\"Asked claude-sonnet-5 to say ready → it said 'ready' in 1.4 s · about 20 tokens, billed to this key\"). When it fails you get one plain sentence and the next step; the HTTP code, the retry time and the provider's own words sit under Details. A rate limit and the test that met it show as one notice per account, and adding an account runs the same test and shows the same sentence. A key the model echoes back is blanked out of its reply, a second press within 10 seconds shows the last result instead of calling the provider again, and an API token can no longer run the test.\n- A Gemini key that is refused Pro models now says why: your Google AI plan covers the Gemini app, not the key, and the key's Google Cloud project has no billing. Pick a Flash model, or enable billing on that project at aistudio.google.com.\n- Lock screen: the buddi version sits quietly in the bottom-right corner (\"buddi 0.1.0-pre.43\"), with \"· update ready\" when a newer buddi is ready to install, the same answer as the status bar.\n- Removing buddi asks the same question everywhere: \"Remove buddi from this Mac?\", in Settings → System and in buddi.app's Uninstall buddi… sheet.\n- Groups: a room whose coordinator isn't buddi can now ask its own members (\"researcher may not ask anchor\" is gone). Being in the room is the permission, for that room only; nothing is added to any agent's settings, each member keeps its own tools and approvals, and the approval card for creating or changing a group says so. And an @mention of a member always reaches it, whether you type its handle, its id or its name.\n- ChatGPT accounts list GPT-6.1 Sol again: buddi now asks ChatGPT for its models as the current Codex version, so newer models are no longer hidden. A new ChatGPT account starts on GPT-5.5, which every plan serves (starting on GPT-6.1 Sol failed on plans without it); when ChatGPT's own list has GPT-6.1 Sol, the picker suggests it, and it's saved when you press Done. Under the model picker you now see where the list came from (\"from ChatGPT · refreshed 3 min ago\" or \"built-in list\"), and Refresh models always asks ChatGPT afresh.\n- Settings → Notifications no longer says \"Telegram is not set up\" under a working Telegram. The line now matches the row: \"Telegram: paired with your phone\", or \"Telegram is running; pair your phone\" with a link when no phone is paired yet.\n- Mail: a mailbox you just connected no longer shows \"No conversations here\". It says \"Connected to you@example.com. Reading new mail from now on; 412 older messages left alone.\" The top of the page names your connected mailboxes, and Mailboxes and rules shows each one as connected, with when it last synced. \"Read the last 7 days\" opens Mailboxes and rules for now, which tells you that reading older mail isn't available yet. Until its first check has finished it says \"Connecting to you@example.com…\" instead, the search bar shows only with the list, and conversations kept from a mailbox you turned off stay listed."
   },
   "files": [
     "packages",