Skip to content
FinityAi logo

Updates

Product highlights, without the noise

Public release notes for customers and evaluators: what changed, why it matters, and where FinityAi is getting easier to trust for delegated work.

Explore featuresView resources

Activity

Release pulse

A day-by-day view of public changelog activity over the last 26 weeks.

Active days

29 / 182

Releases logged

29

Last release

15 days ago

LessMore

Latest

Recent public highlights

Short summaries only: user-visible improvements, clearer controls, and connected-work updates.

optional updates

  • optional Flutter sample app in the repository for iOS and Android assistant chat while signed in as you.
  • saves can detect when a note changed on another device so you can use the server copy, overwrite with yours, or keep editing; unsaved drafts on this device can be recovered after a reload.
  • select multiple notes to archive, pin, restore, or delete (archived), tune search to titles only or full note text, and download a full backup zip from Integrations.
  • removing one archived Personal Note is now confirmed in chat before it is deleted.

when updates

  • when a day has more events than fit on the month grid, the full list from “more” is ordered by start time so you can scan the day in order.
  • Free plan accounts can now save up to 10 schedules instead of 3.

starting updates

  • starting another new chat from a project hub no longer reopens your previous thread. New chat from the sidebar while you are viewing a project stays in that project.
  • the header breadcrumb on a project hub shows your project name instead of generic “P” and “Details”.
  • when document uploads are enabled, attached files can route through a document-aware model path (with fallback) so file questions are less likely to fail silently.
  • Assistant model routing now defaults to using any configured provider (instead of OpenAI-only), and the Anthropic auto “agent” tier now uses Sonnet for stronger tool-use quality while nano/fast remain on Haiku.

Archive

Earlier milestones

Questions about a change?

  1. starting updates

    • starting a new chat from a project no longer briefly opens a thread and then drops you on the empty assistant home.
    • opening or starting a chat from the projects area uses the same full assistant chat experience as the main Assistant screen, with one consistent address bar pattern for threads.

  2. chat updates

    • chat lists and Search chats stay closer to what’s on the server after you send a message, rename a chat, or move it between projects.
    • after your first message, the chat stays on the thread instead of flashing “loading” and jumping back to the empty home screen.
    • turning on deeper reasoning before you send the first message now applies on that first reply.
    • links that open a specific section (for example Notifications or Billing) now reliably show that section instead of the wrong one.

  3. Messaging updates

    • Messaging (Telegram, Slack, Signal, WhatsApp, Discord): send /new, /reset, or phrases like “fresh chat” to clear the bot thread and start over.
    • uses your usual “tools & integrations” setting so connected apps work instead of replies acting like access was blocked.
    • Next actions sits over the right edge of the thread so your messages stay centered like the main chat instead of sliding when tools run.
    • changing the model now updates the same account-wide assistant preference as the main chat, so the footer label and other sessions stay in sync.

  4. creating updates

    • creating a project or switching projects should feel steadier, with fewer flashes and fewer mistaken redirects when the connection is flaky.
    • opening a project from a shared or bookmarked link should land on the project hub instead of briefly showing an older chat.
    • New project and Create project on the main Projects screen should open the create dialog again.
    • when you open a project from a link, the first-load message now says you’re loading the project instead of the whole conversation list.

  5. search updates

    • search events with ⌘K (Ctrl+K on Windows)—filter the grid by title or description and jump to an event’s date.
    • your saved assistant chats stay yours—only you can open them, not teammates or other accounts.
    • your first message on a new chat is less likely to stall—the conversation link updates once the server confirms it, so sending right away works more consistently.
    • Search chats opens the chat picker from Calendar, Tasks, Notes, and other app pages instead of jumping to Assistant.

  6. after updates

    • after quick connect, the Finity bot sends the same short confirmation in Telegram as the code link flow when Telegram allows delivery (open the bot once if you do not see it).
    • after one-tap connect, the Finity bot sends the same short confirmation in Discord as the code link flow when it can open your DMs.
    • opening the main Projects page shows your project list instead of an old chat that was still selected in the background.
    • starting a new chat right before the conversation list finishes loading no longer jumps you to an older thread until the assistant replies.

  7. open updates

    • open Projects from the sidebar — browse all projects, jump into a dedicated workspace per project, and keep chats organized without mixing them with the rest of the app.
    • open a matching folder in Personal Documents so files you add there stay with that project; chats inside the project can search that folder first.
    • group browser chats into folders from the sidebar — filter the list, create folders, rename or delete them; new chats follow the folder you’re viewing.
    • your time zone lives under User settings → Profile (we suggest your browser’s zone). That keeps Microsoft 365 imports and your calendar aligned with your local days.

  8. when updates

    • when your workspace has open models enabled, you can pick them in the model menu alongside the other cloud providers.
    • rename a web chat from the thread menu in the sidebar (⋯) or with the pencil next to the title at the top; names stick when your threads are saved to your account. Chats mirrored from messaging apps stay labeled by channel here.
    • on the Telegram setup page, you can email your account a link to open the Finity bot on your phone if you do not have Telegram on the computer you are using.
    • link Discord to your Finity account in one step when your workspace enables Discord sign-in; you can still use the manual user ID and bot code path if you prefer.

  9. describe updates

    • describe a small interactive web app in plain language, preview it safely in your browser, save versions, and reopen them later.
    • connect with Quick connect or a personal API token, see status on the integrations gallery, and work with boards and items from the assistant when Tools are on.
    • connect with Quick connect (OAuth) first, or use a personal access token fallback, then confirm connection health and use assistant tools to list bases/tables plus read or update records when Tools are on.
    • stock cards no longer default to US$0.00 when the data feed returns placeholder zero values—price now falls back to prior-close math or recent close history, and zero-only day bars are hidden instead of shown as real values.

  10. when updates

    • when the assistant bridge hits an error, the same message is no longer repeated to your DM every few seconds—brief cooldown matches Telegram and Discord behavior.
    • assistant replies sent to Telegram, Signal, Slack, and Discord are formatted for each app—Telegram shows proper bold and lists instead of raw markdown symbols, and other channels strip or adjust markup so messages read cleanly.
    • saving or creating a runable now requires task instructions—you can’t save an empty runable while the schedule is off anymore.
    • linking accepts the confirmation code even if you paste it with backticks, and the page rechecks until your account shows as linked.

  11. a updates

    • a new Playground (beta) is available at Assistant Playground, where you can test prompts, model choices, and tool behavior in one lab view with request/stream inspection details.
    • Integrations → Market data now includes a dedicated page with what the tool can do and sample prompts, so it is easier to understand stock snapshots before use.
    • Integrations → Discord is now live for DM chat: link your Discord user ID once, then continue chats from Discord with the same account, assistant settings, and enabled tools as web chat.
    • saved daily and weekly schedules now run automatically, instead of requiring manual Run now every time.

  12. when updates

    • when your message draft isn’t empty, Improve prompt picks up a gentle motion hint so it’s easier to spot (honors reduced-motion preferences).
    • while Improve prompt is running, you can’t send the message (keyboard, send button, and attachments stay locked until it finishes).
    • Deeper reasoning tooltip copy uses colors meant for the primary tooltip surface so it stays readable in dark mode (no page “muted” text on the colored bubble).
    • header logo alignment—on the collapsed icon rail, horizontal padding matches the nav so the logo lines up with icons below; when expanded, the wordmark sits vertically centered with the title and subtitle block.

  13. personalization updates

    • personalization controls (base style, tone sliders, emoji preference, custom instructions, and optional “about you” context)—saved to your account and applied to assistant replies on web and mirrored channels.
    • each schedule shows when the next run falls (relative time like “in 2 hours”, plus the next local date/time in your chosen zone). In the editor you also see the current clock in that zone for comparison.
    • Run now runs the saved task immediately with the same assistant and tools as a scheduled execution (from each card and from edit). Telegram delivery follows your saved delivery setting when linked; manual runs do not replace or cancel the next scheduled slot for that day.
    • Finity Workspace user guide (/app/docs/personal-apps)—the suite navigation strip (Tasks · Notes · Calendar · Documents · Runables), turning apps on from Integrations, how Finity Workspace differs from organization integrations, per-app overview with deep links, Finity Runables pointer, Assistant context, and troubleshooting.

  14. Documentation updates

    • Documentation → Microsoft 365 is now available inside the app (/app/docs) with setup steps, mail vs calendar guidance, sample prompts, troubleshooting, and links to official Microsoft references.
    • when a search misses the query input, the error now explains what kind of text to provide (for example contact name/email, company name/domain, or ticket subject) instead of only saying "query is required."
    • tool rows under Behind the answer now include richer, safer context across all tools (like limits, counts, object types, and toggles) while still hiding sensitive values.

  15. calendar updates

    • calendar actions are now available in chat for connected accounts, including listing calendars/events plus create, update, and delete event flows.
    • expanded support for Teams and OneDrive actions, including team/channel reads, Teams meeting helpers, and OneDrive file/folder operations.
    • Behind the answer now uses clearer, task-level tool copy for more integrations so each tool row is easier to understand at a glance.
    • prompt-improve behavior now has a dedicated API route path, making refinement actions more consistent with normal chat send flows.

  16. sending updates

    • sending a message while your thread list was still loading no longer briefly jumps you to a different chat and back when the reply arrives.
    • when you ask for a U.S. stock price (with Stocks enabled), the reply includes a snapshot card with price, day range, optional bid/ask, a simple chart, and range toggles—alongside the written answer.
    • when you use Auto routing, each reply’s small model line shows a simple band (Auto · Lite, Standard, or Premium) instead of the raw model name; hover it to see which model ran.
    • we removed the extra status bar and Wake GPU controls for local models—the chat area stays simpler; you’ll still see normal errors if a request fails.

  17. when updates

    • when connected, the setup page shows your Notion workspace (and integration name when Notion provides it) so you can confirm the right workspace.
    • when the model is Auto and you use Finity-hosted models, usage counts toward your hosted token allowance; choose a specific model when you want that model’s billing path.
    • choose Simple page read (fast open-link style), Browser Use (agentic browsing), or both.

  18. when updates

    • when the model server banner appears, use Check status to probe again instead of waiting for the automatic poll.
    • Wake GPU always shows a short confirmation (before, success with no toast looked like nothing happened).
    • Auto can route each message to the best-fit fast or premium model so simple questions stay inexpensive and harder work gets a stronger model.
    • when a reply uses Auto, the footnote shows the resolved model and a short task hint; hover for routing details (token counts stay on the same line when available).

  19. longer updates

    • longer tasks (partner lists, multi-page extractions) now finish without the chat timing out with "No reply text was returned". The browser tool also runs faster and gathers more results before giving up.
    • routed back to OpenAI by default so tool-using replies work reliably again (listing your notes, tasks, calendar, etc.).

  20. richer updates

    • richer create/query/update (row fields, initial blocks, database patch, single-property reads); search can sort by last edited time; append supports after; notion_get_me and file-upload lookup; re-fetch a block to refresh inline file download links.
    • the assistant can read and search Excel workbooks (.xlsx)—sheets are turned into plain text (large sheets may be truncated).
    • create page explains page vs database parents, requires parent_id when the server has no default, and surfaces Notion’s own error text when creation fails—so mismatched parent types (a common HTTP 400 cause) are easier to fix.
    • creating a database row now detects the database’s real title column name (for example “Project name” instead of default “Name”), and the assistant avoids repeating the exact same failed Notion tool call in one turn.

  21. describe updates

    • describe your scheduled task, use Improve to make it more AI-friendly, then set when it runs and how you get the alert (e.g. Telegram).
    • on wide screens, a Next actions panel appears beside the chat only when that reply actually runs tools; on smaller widths, tool rows sit above reasoning so they aren’t buried under a long thinking block.
    • if a reply uses several tools in one step, every action appears in Next actions (and the Tools strip) at once, then each row updates as it finishes.
    • the assistant now sends date ranges the way Triple Whale’s API expects, so requests like a week of revenue should succeed when your key and store are set up.

  22. optional updates

    • optional OpenAI hosted web search (Responses API) alongside Browser Use—uses your OpenAI models without turning Tools on.
    • if you connected HubSpot with Quick connect (OAuth), the assistant now receives that connection for CRM tools—same as a Private App token.
    • web reading and multi-step browsing (Browser Use) now run on a dedicated hosted browser session, separate from the chat API.
    • Quick connect opens HubSpot sign-in in a separate window so you stay on the HubSpot integration page; if your browser blocks popups, the same sign-in flow still opens in this tab.

  23. the updates

    • the tools-off reminder is easier to read (neutral styling, one short explanation, same steps to enable Tools and link apps).
    • message footers no longer show “tokens n/a” when the model does not return usage; time and latency still appear when available.
    • the chat screen becomes interactive sooner after you open it; the app refreshes model and connection status in the background instead of waiting on every load.
    • when you return to the same chat, message history should appear sooner—the app loads your thread and the conversation list at the same time instead of one after the other.

  24. new updates

    • new public blog with short product notes (starting with hosted execution vs chat-only tools).
    • saving in the edit drawer should no longer flash or jump the list the way it sometimes did when the row under the panel caught a stray click.
    • new post on crowdfunding deal flow (weekly scans, Finity notes, reminders before close) with a sample assistant prompt you can adapt.
    • five new workflow stories (WhatsApp, Slack, email and Notion, everyday one-line asks, and one assistant across channels), each with a copy-paste assistant prompt.

  25. connect updates

    • connect your phone under Integrations → WhatsApp, copy the link code FinityAi shows, then send that code as your only message from WhatsApp to your organization’s FinityAi business number. After linking, chat with the assistant from WhatsApp like other channels.
    • connect your Google account under Integrations to give the assistant tools for Gmail, Google Calendar, and read-only access to files in Google Drive (including exporting Google Docs and Sheets as text).
    • the Dashboard, Documents, and Personal Documents integration pages show how much of your personal storage you’re using, with a 1 GB cap on the free tier (higher tiers follow your workspace plan). Uploads are blocked when you’re over the limit.
    • the assistant can read body text from Word files (.docx) in addition to plain text and common text/code types.

  26. We updates

    • We added a Triple Whale integration (ecommerce analytics): save an API key under Integrations, then ask the assistant for summary metrics or attribution exports when tools are on.
    • This page—what’s new, what we fixed, and known issues—for people signed into the app.
    • If you ask the assistant for something we don’t support yet, it may offer to log a request for you (with your OK).
    • the assistant can search your files, read parts of files, get file info, and suggest changes (add, edit, delete, move, new folder). A Confirm/Dismiss banner appears in chat—nothing is applied until you confirm.
Looking for setup help? Start with the documentation or the FAQ.
Changelog — FinityAi · FinityAi