Latest updates
New updates and improvements to allmates.ai. Bigger releases get their own announcement.
Last updated About 1 hour ago
August 2026
🧰 Snowflake, four ways
Added · Tools & Integrations · August 28, 2026
Four Snowflake tools join the store, so a Mate can work directly against the warehouse your company already runs on. Cortex Analyst answers a plain-language question and shows the SQL it used. Cortex Search finds documents by meaning rather than keyword. Cortex Agents return an answer with its reasoning and sources. And Snowflake SQL runs queries directly for schema exploration.
💬 No more silent pause when a Mate answers without being mentioned
Fixed · Chat & Collaboration · August 28, 2026
When you send a message without mentioning anyone, allmates.ai works out which Mate should pick it up. That resolution used to happen in a first pass that produced nothing visible, so the conversation sat silent for a moment before the answer began. The Mate now starts responding on the first pass, and you see it working straight away.
📊 Suspended members no longer distort your engagement numbers
Improved · Analytics · August 27, 2026
Members you have suspended were still counted in the inactive population of the Engagement dashboard, which inflated the inactive figure and pulled the activation rate down. They are now excluded from that population. Their past activity is untouched: if someone was active in June and suspended in August, June still counts in your active-user KPIs, rankings, cohorts, and costs.
🧰 Adobe Analytics and Customer Journey Analytics
Added · Tools & Integrations · August 14, 2026
Two new tools connect your Mates to Adobe's analytics stack. Explore report suites and data views, run ranked and trended reports, and manage segments and audiences — from a conversation, without leaving for the Adobe interface to pull a number.
🎙️ Know your note taker got in, while it still matters
Added · Meeting Notes · August 2, 2026
The browser extension now checks the meeting link before sending the note taker, and warns you if it looks like the bot will be refused. Once dispatched, the meeting page itself shows the bot's real state instead of leaving you guessing, and escalates to a visible alert when it cannot get in. You no longer discover a failed join ten minutes later, in a panel you were not looking at.
✨ A serious performance pass on the mobile app
Improved · User Experience · August 2026
Opening a conversation now puts you straight on the latest message, anchored where you expect, instead of watching the page glide down while everything settles. While a Mate streams its answer, only that message is recomputed — the rest of the thread stays still, which is what long conversations on a phone were waiting for. Heavy content that used to fight the stream, code blocks, tables and document widgets, renders without stalling it. The installed app also updates cleanly in one pass rather than reloading itself twice, and the sidebar, safe areas and text wrapping were gone over on real devices.
🧰 Bring the open web into a conversation
Added · Tools & Integrations · August 2026
Two tools for working with pages that live outside allmates.ai. Exa searches the web with a neural engine and hands back clean, readable content instead of raw HTML. The web scraper extracts pages you name — public ones, or ones behind a login — as structured Markdown, so a Mate can work from a supplier portal or an internal wiki the same way it works from a document you uploaded.
🧰 The tool store now covers most of where your work already lives
Improved · Tools & Integrations · August 2026
The catalog has grown to around forty connections. Your Mates can work in Slack, Notion, GitHub, Atlassian, Monday, Miro, Canva, HubSpot, Qonto, Qlik, SurveyMonkey and Legifrance, across the Microsoft 365 suite — Outlook, Teams, SharePoint, OneDrive, Excel, Word and PowerPoint — and against Snowflake and Adobe Analytics. Each one connects from the store, with credentials held once for the organization rather than set up by every person who needs them.
💬 Answers that come back as documents and diagrams, not just text
Added · Chat & Collaboration · August 2026
A Mate can now return a fully formed HTML widget inside its answer — multi-column layouts, tabs, animation, real typography — rendered in an isolated sandbox next to the conversation, ready to copy or export to PDF. It can also return an SVG widget, which is genuine SVG source rather than a generated image, so a diagram stays sharp at any size and remains editable. The two are separate output types: use HTML for a structured overview, SVG to explain one technical idea precisely.
✨ A chat that fits a phone screen
Improved · User Experience · August 2026
The mobile conversation was rebuilt around the width it actually has: messages run edge to edge instead of wasting margins, message actions moved into a menu rather than crowding the bubble, and the composer expands as you write instead of trapping a long message in two lines.
July 2026
🎙️ A more dependable note taker, end to end
Improved · Meeting Notes · July 24, 2026
A batch of improvements to meeting recordings. Long recordings now transcribe from start to finish, however long the meeting ran. Summaries and emails come back in the language actually spoken in the meeting, instead of being guessed from the opening exchanges. When a recording fails to import, Retry now recovers it rather than leaving it stuck. And the extension keeps your session alive, so sending the note taker works on the first try.
🎙️ Work with a recording, not just listen to it
Added · Meeting Notes · July 2026
Recordings get a full page of their own. The transcript follows the player as it plays, search filters it down to the passages you want, and you can correct it before anyone else sees it — including find and replace — with your edits kept. From the list, send the summary and transcript by email to people who were not in the room; the completion email now carries the full transcript as a file and a chart of who spoke for how long. And when several people ask for a note taker on the same meeting, one bot goes in and everyone gets the recording.
🤖 Set how hard a Mate thinks, and how freely it writes
Added · Mates · July 2026
Two sliders on a Mate's configuration. Thinking effort runs from none through to maximum and follows what the chosen model is actually capable of — turn it up for a problem worth reasoning about, down for a quick answer that costs less. It is marked Beta while we learn how people use it. The second slider runs from rigorous to creative, within the range the model supports, so you decide whether a Mate sticks closely to the facts or is allowed to range.
📚 A working folder inside the conversation
Added · Knowledge · July 2026
A conversation can now hold files the way a project folder does. A Mate reads them, writes to them, and keeps them across turns, and anything it produces there can be pulled back out as an attachment you download. Spreadsheets generated by a tool land in the same place, referenced rather than re-pasted, so a long analysis stops being a chain of copies.
🧠 Models hosted on Microsoft Foundry
Added · AI Models · July 2026
Organizations that run their models through Microsoft Foundry can now route chat traffic there. It matters when your agreement with Microsoft, your data residency, or your procurement rules say model calls stay inside your Azure tenancy.
🧠 Steer the reasoning on more than one provider
Improved · AI Models · July 2026
Reasoning effort used to be something only some models understood. It now applies across OpenAI, Azure and OpenRouter as well, each model declaring what it actually supports so the control offers the levels that model can honour and nothing else. The GPT-5 family declares its reasoning capabilities in the catalog alongside the rest.
📊 Cache savings show up in what you are billed
Improved · Analytics · July 2026
When a long conversation reuses cached context instead of resending it, those tokens cost less than fresh ones. They are now counted separately and priced at their real rate, so your usage figures reflect the saving rather than charging cached context as though it were new.
💬 Mathematical notation that reads like mathematics
Added · Chat & Collaboration · July 2026
Formulas used to arrive as raw LaTeX, which is fine if you can read it in your head and unhelpful otherwise. Mathematical notation now renders properly in chat — fractions, integrals, matrices, Greek letters — inline within a sentence or displayed on its own line, so a Mate explaining a calculation can actually show it.
✨ A proper app on your phone
Improved · User Experience · July 2026
allmates.ai installs on a phone as an app, and now tells you when a new version is ready instead of quietly serving you an old one. Alongside that, a pass over the mobile experience: mention and emoji suggestions no longer shove the layout around above the keyboard, content respects the safe areas on notched screens, and long text wraps where it should.
🧰 Code that runs, and files that stay put
Added · Tools & Integrations · July 2026
A Mate can run code in an isolated sandbox to compute, transform a dataset, or produce a file. The files it creates now survive from one turn to the next in the same conversation, and connect to the files already attached to that chat — so you can generate a spreadsheet in one message, refine it in the next, and download it at the end, instead of starting from nothing each time.
June 2026
🧠 Claude Fable 5 is available
Added · AI Models · June 10, 2026
Anthropic's first publicly available Mythos-family model joins the catalog, with advanced reasoning and vision, at premium pricing.
🤖 Put a Mate on a schedule
Added · Mates · June 2026
Some work does not need you there to ask for it: a Monday morning summary, a weekly report, a monthly review. A Mate can now run on its own — daily, weekly, or monthly — targeting a workspace or a conversation, and its answer waits for you when you arrive. Scheduled conversations are marked as such, each run leaves a card with its history, and the people who should read the result are notified when it lands.
🎙️ Sharper recordings, sharper summaries
Improved · Meeting Notes · June 2026
A broad pass over meeting capture. Your recording history shows all of it, not just the most recent entries. Summary emails name the people who actually spoke, drawn from the meeting roster instead of guessing. Language detection reads the whole conversation rather than its opening seconds, and a near-empty transcript no longer invents content. Audio-only recording captures system sound, local video recording handles more camera and screen setups, and Teams links that require sign-in are handled properly.
🤖 Decide what each Mate is allowed to do
Added · Mates · June 2026
A Mate's capabilities are now edited in one place, with guidance explaining what each setting changes: which widgets it may produce, how it handles context, what generation options it exposes. Settings are drafted and reviewed before they take effect, so tuning a Mate that people rely on daily no longer means experimenting in production. Generated avatars also moved to a newer image model and come back sharper.
🧰 SurveyMonkey, and SharePoint sites you choose
Added · Tools & Integrations · June 2026
SurveyMonkey joins the store, so a Mate can pull survey results into an analysis instead of waiting for someone to export a spreadsheet. On the Microsoft 365 side, the SharePoint connection can now be pointed at selected sites rather than the whole tenant — useful when only a few document libraries should ever be in reach.
🗂️ Folders, for workspaces that outgrew a flat list
Added · Workspaces & Organization · June 2026
A workspace that accumulates hundreds of conversations becomes a scrolling exercise. Workspaces can now hold shared folders, nested as deep as you need, shown in the sidebar and on the workspace page. Drag a conversation into a folder, or move it from the dialog. Admins create, rename, reorganise and delete folders; deleting one moves its contents up a level rather than taking the conversations with it.
🎙️ Delete a recording and have it stay deleted
Added · Meeting Notes · June 2026
Deleting a recording now removes everything behind it: the history entry, the transcript, and the stored video file. Nothing lingers on our servers waiting to reappear, which matters when a meeting should never have been recorded in the first place, or when a retention policy says it is time for it to go.
🧠 Grok, connected directly
Added · AI Models · June 2026
xAI's models are now reached through a native connection rather than a third-party gateway. Fewer hops between your question and the answer, your own xAI credentials, and the model's capabilities declared properly in the catalog instead of inherited from whatever the intermediary reported.
🎙️ Find out the moment a note taker cannot join
Added · Meeting Notes · June 2026
A note taker that fails to get into a meeting now tells you right away, through a browser notification and an email, rather than sitting quietly in a panel you are not watching. The message carries the actual reason it was turned away, so you know whether to fix the link, adjust the meeting's admission settings, or simply record it yourself.
May 2026
🧠 Claude Opus 4.8 becomes the new reference
Added · AI Models · May 29, 2026
Sharper judgment and better document analysis than 4.7, at unchanged pricing. Worth switching your Mates over.
🧠 Gemini 3.5 Flash joins the catalog
Added · AI Models · May 20, 2026
Cost-effective reasoning with adaptive thinking and a million-token context window, for Mates that need speed without giving up depth.
🤖 Tell the official Mates apart at a glance
Added · Mates · May 11, 2026
When an organization has many Mates, several can look alike or seem to do the same job, and there was no way to tell which one your company actually endorses. Admins can now mark a Mate as verified. The badge shows on the Mate card, and the list can be filtered to show only verified Mates — so people land on the curated agent rather than a lookalike.
🎙️ Send a note taker into your meetings
Added · Meeting Notes · May 1, 2026
Paste a meeting link and allmates.ai sends its note taker in for you. Recording happens entirely on our side, so you can close the tab, close your laptop, or not attend at all — it keeps going. Google Meet and Microsoft Teams are supported, for meetings up to three hours. The recording lands in your history alongside the ones you capture yourself, with the same transcript and playback.
🧠 Grok can call your OpenAPI tools
Fixed · AI Models · May 2026
Mates running on Grok could not use tools defined from an OpenAPI schema — the call was rejected before it ever reached your service. Grok now works with the same REST tools as every other model in the catalog, so the choice of model no longer restricts the choice of tools.
💬 Copy or move a conversation, keep all of it
Improved · Chat & Collaboration · May 2026
Moving a conversation into another workspace, or copying it to branch off, now brings everything along: attachments, tool call logs, and the original ordering of messages. What you find on the other side is the conversation you had, not an approximation of it.
📊 Model costs you can rely on
Improved · Analytics · May 2026
Cost figures in the analytics dashboards are now computed from the model price table, so each model's real rate is applied rather than an approximation. Comparing what an expensive model costs against a cheaper one, per team or per Mate, gives you a number you can act on.
⚙️ Microsoft SSO matches the right person
Improved · Administration · May 2026
In many Microsoft tenants a person's sign-in name and their primary email address are not the same string, which used to be enough to keep them out or land them in the wrong account. Matching now handles that case. While we were there, the member picker for private workspace access shows first and last names, so you pick a colleague rather than an address.
April 2026
🧠 OpenAI Image 2.0 for image generation
Added · AI Models · April 29, 2026
High-resolution generation with reliable multilingual text rendering and stronger visual reasoning.
🧠 GPT-5.5 is available
Added · AI Models · April 27, 2026
OpenAI's flagship joins the catalog, matching and in places surpassing Claude Opus 4.7.
🧠 Model retirements you can see coming
Improved · AI Models · April 22, 2026
Model providers retire their models on their own schedule. A model on its way out is now flagged as deprecated in the catalog, so you can see it before it disappears and move your Mates to a successor on your own terms, instead of finding out when a conversation stops working.
🧠 Your Mates keep answering when a model goes down
Improved · AI Models · April 21, 2026
When a provider is overloaded or down, allmates.ai switches to a fallback model and carries on instead of handing you an error.
⚙️ Invite your team with a single link
Added · Administration · April 18, 2026
Bringing people in no longer means sending invitations one address at a time. Generate an invite link for your organization and share it wherever your team already talks — a channel, an onboarding document, an email. Anyone who follows it joins your organization directly.
🧠 Claude Opus 4.7 raises the document ceiling
Added · AI Models · April 16, 2026
Document processing up to 600 pages, with stronger long-context reasoning and high-resolution vision.
🧰 Five times more tools, connected in one click
Added · Tools & Integrations · April 14, 2026
A wave of new MCP servers multiplies the tool library by five, with OAuth-secured connections to enterprise applications.
🤖 Take your Mates with you
Added · Mates · April 1, 2026
Export a Mate's configuration as a file to back it up, hand it to another organization, or rebuild the same setup elsewhere.
🧰 Tools that know who is calling them
Added · Tools & Integrations · April 2026
A tool or MCP server had no standard way of knowing who was on the other end of the call. Tool parameters now accept placeholders such as {{member.fullName}} or {{member.id}}, substituted with real values just before the call runs. Your integrations can personalise an answer, filter on the caller, or write back under the right identity, without anyone having to type that information into the conversation.
💬 Choose what context a conversation inherits
Added · Chat & Collaboration · April 2026
Context cascades from your organization down to the workspace and into each conversation, which is usually what you want — and occasionally exactly what you don't. A workspace can now opt out of the organization context entirely, which is what you need for a sandbox or an isolated project. And when you open a chat, you can decide whether it picks up the workspace and organization context at all, so a quick exploratory exchange starts light instead of carrying everything. Personal workspaces follow the same rules, and workspaces created before the setting existed keep inheriting as they always did.
⚙️ Point Anthropic connections at your own endpoint
Added · Administration · April 2026
An Anthropic connection can now carry its own base URL. If your organization routes model traffic through a corporate gateway, a regional endpoint, or your own proxy for logging and compliance, you configure it on the connection instead of accepting the default route.
🤖 Find the right Mate in a growing roster
Improved · Mates · April 2026
Once an organization runs dozens of Mates, scrolling stops working. The list was reworked around filters that combine categories, models, and tools, with grouped category navigation and inline actions so you can act on a Mate without opening it. Mates can also be renamed, so a name chosen in haste is no longer permanent, and reinstalling a Mate from the store updates the one you already have instead of leaving you with two.
✨ Faster answers, and a lighter app
Improved · User Experience · April 2026
A sustained push on speed and cost. Mates now start answering far quicker after a quiet period — cold starts dropped by nearly three quarters — and invocation is faster across the board. Long conversations reuse cached context instead of resending it, which cuts what they cost in input tokens. Attachment storage was reworked, and the app itself is lighter to load: translations fetched on demand, fonts served from our own domain, connections opened ahead of time.
💬 A steadier conversation
Improved · Chat & Collaboration · April 2026
A run of refinements across the chat. Mentioning a Mate selects that Mate, including when you start a fresh conversation. Re-mentioning one that stopped early no longer produces two answers at once. Charts with detailed options — custom axes, titles, legends — render instead of blanking the message they belong to. Histories involving several tools running in parallel are rebuilt in the right order, and global search now finds all of your personal projects.
💬 Share a conversation with a public link
Added · Chat & Collaboration · April 2026
A conversation worth showing to someone outside allmates.ai can now be shared as a public link. What you share is an isolated snapshot, attachments included, so the original conversation stays private and keeps evolving without changing what your reader sees. Set an expiry when you create the link, revoke it whenever you want, and see how many people opened it.
🧰 Tools that carry their own credentials
Added · Tools & Integrations · April 2026
A tool connection now holds its own authentication, encrypted and shared across everyone in the organization, so a tool is set up once rather than by each person who needs it. OAuth connectors attach to a specific connection, and a single Mate can now mix MCP servers and REST API tools instead of having to choose between the two.
📚 Decide how each document reaches your Mate
Added · Knowledge · April 2026
Documents used to be handed to a Mate one way, whatever they were. You can now pick the integration mode per attachment: the default handling, a Markdown conversion, a summary, or a reference the Mate consults only when it needs to. A hidden mode also lets a document inform the Mate without appearing in the conversation. A 200-page contract and a one-page brief no longer have to cost the same context, and each document now shows what it costs in tokens so you can see where your context budget actually goes.
March 2026
🤖 Clone a Mate in one click
Added · Mates · March 31, 2026
Duplicate a Mate into an independent copy with the same configuration, to experiment or cross-validate without touching the original.
📊 Look at exactly the period you care about
Added · Analytics · March 2026
Analytics dashboards were limited to a handful of preset ranges, which never quite matched the quarter you were reporting on or the fortnight after a rollout. You can now set an arbitrary start and end date, and every chart, KPI and ranking follows.
🤖 Set up a whole team of Mates at once
Added · Mates · March 2026
Standing up an organization used to mean creating Mates one at a time. A guided multi-step import now brings in a whole set in one pass — useful when you are rolling allmates.ai out to a new department, or replicating a configuration that already works somewhere else.
⚙️ Provision users without doing it one by one
Added · Administration · March 2026
An administration page for creating and updating members in bulk, backed by an API for teams that would rather drive it from their own tooling. Onboarding a department, applying a reorganisation, or correcting a batch of accounts stops being an afternoon of clicking.
✨ Better on a phone, and on the very first visit
Improved · User Experience · March 2026
Three things that made a bad first impression, now sorted. The message box stays where you can reach it on a phone instead of slipping under the keyboard, workspace navigation is reachable on small screens, and opening allmates.ai for the first time no longer stalls on a blank screen while everything loads.
⚙️ Authorizations that hold up in a large organization
Improved · Administration · March 2026
The authorizations table resolves members on the server and pages through them, instead of loading everyone into the browser at once. Granting or reviewing access stays responsive whether your organization has thirty people or several thousand, and a search field on the Mates and tools authorization tabs finds the one you are looking for without scrolling.
🗂️ Switch a workspace between private and public
Added · Workspaces & Organization · March 2026
A workspace used to be locked into whatever visibility it was created with — a private space that grew into a team effort had to be rebuilt from scratch to open it up. Admins can now flip the visibility of an existing workspace either way, and its conversations follow.
February 2026
💬 See that a Mate is working, not wonder
Added · Chat & Collaboration · February 2026
A Mate that is thinking, reading a document, or calling a tool now shows it, instead of leaving you in front of a still screen deciding whether to send the message again. Conversation names are also generated in the background, so a new chat opens immediately rather than waiting on a title nobody asked for.
✨ Talk instead of typing, and send the way you prefer
Improved · User Experience · February 2026
Dictation in the browser extension works again, for the times when saying it is faster than writing it. And the shortcut that sends a message is no longer fixed: an organization can set the default its people expect, so nobody keeps sending half-written messages because Enter did something different in their last tool.
✨ allmates.ai in your language
Added · User Experience · February 2026
The interface is available in English, French, Spanish, Italian, Portuguese, Japanese and Swedish — and localization now reaches past the buttons, covering the prompts, tool descriptions and model labels people actually read while working. A team spread across countries stops having to agree on one working language just to use the product.
✨ Replay a conversation instead of scrolling through it
Added · User Experience · February 2026
A good conversation with a Mate is often worth showing to someone else — a colleague, a client, a room. Replay plays it back message by message, formatting and all, so you can present the reasoning as it unfolded rather than scrolling a wall of text. It works on shared conversations too, for people who have no account.
✨ Get your content out of a conversation
Added · User Experience · February 2026
What a Mate produces often needs to live somewhere else. You can now copy or export from a Document widget, from a single message, or from an entire conversation — including charts and code blocks. Going the other way, pasting a long block of text turns it into a Markdown attachment rather than flooding the composer.
January 2026
⚙️ Take the app down gracefully, and say why
Added · Administration · January 2026
Maintenance no longer means users hitting errors and wondering what happened. Admins can put the app into maintenance mode, and post a banner message that everyone sees on their way in — to announce a window in advance, explain an ongoing one, or flag anything the whole organization should know.