Changelog

• New Ask Titan AI about your Compass setup in plain words, for example "Show my Compass suppliers and their lead times" or "Which of my SKUs are not fully set up in Compass". It lists the suppliers in your supplier book with their lead, freight and buffer times, total landed time, payment split and balance terms, and shows each SKU's Compass settings on each sales channel: whether it is fully set up, which supplier it is linked to, when freight is paid, and its Demand Plan ordering settings.
• New Titan AI can also make changes in Compass for you. It can add a supplier or update a supplier's times and payment terms, delete a supplier, and change a SKU's linked supplier, when freight is paid, and Demand Plan settings such as minimum stock days, carton size and MOQ. Each change goes through the same approval prompt as Titan AI's other actions. Updating a supplier also updates every SKU linked to it. A deleted supplier cannot be restored, and the SKUs that were linked to it keep its last terms until you link them to another supplier. A SKU can be moved to another supplier but not unlinked.
• Note A change made from chat does not recalculate your forecast, Demand Plan or Purchase Orders. Open Compass and accept the recompute prompt before relying on those figures.

• Fixed Asking Titan AI which search terms to negate could come back with no recommendations on a search-term report longer than about 50 search terms. Titan AI reads the whole report page by page, and it read it sorted by spend. Many search terms share exactly the same spend, so rows could swap places between pages while it read: one search term came back twice and another was never read at all. Titan AI does not recommend negations from a report it has not fully read, so it stopped. It now reads the report in search-term order, which does not shift between pages. If a read still comes back inconsistent, Titan AI still does not recommend from part of the report.
• Fixed When you asked a connector such as Higgsfield to work from a picture you attached in the chat, it could be handed a different picture from the one you meant. Titan AI referred to your attached pictures by their position in the conversation, so the same picture could be referred to differently from one message to the next. Each picture you attach now keeps one name for the whole conversation, including pictures you attached on an earlier message.

• Fixed When Titan AI stops to ask you something and the task then ends before you answer, the question no longer reappears as if it were still waiting. Opening the chat again used to bring the question back, but the task behind it had already finished, so choosing an answer or typing your own made the question disappear and nothing happened. Reloading brought it back and the same thing happened again. The transcript now shows that question as Not answered, and if you do answer a question that has closed, Titan AI tells you that your answer could not be delivered and asks you to type your reply in the chat box, which continues the conversation as a new message. A question that is genuinely still waiting for you is unchanged, including after a reload.
• Fixed Asking Titan AI in chat to add keywords to the Keyword Rank Tracker could come back with nothing added and a note that a marketplace was required, even though the chat already knew which store it was working on. The same could happen when creating a Keyword Relevancy dataset or adding and removing ASINs on one. These actions now go to the store the chat is working on without you having to name a marketplace, and naming one does not change where they land. Reading keyword ranks by marketplace works as before.
• Improved Deleting a custom connector that signs in with OAuth now also asks the sign-in provider to revoke Titan AI's access, where the provider offers a way to do that. The message after you delete says which happened: the access was revoked, the provider offers no way to revoke it from here, or the revocation did not go through. In either of the last two cases, remove Titan AI from that provider's own list of connected apps to end the access now. The connector is always removed from Titan AI, whatever the provider answers.

• Fixed Some masterclass replays feature an invited guest presenting their own company's services. When an answer draws on one of those, Titan AI now attributes the offer to the guest's company, and names the company when the lesson does, instead of presenting it as something Titan provides. This applies in chat and in Titan AI MCP.
• Fixed In chat, asking for your ad spend broken down by match type now keeps Sponsored Products apart from your other ad types. Spend from Sponsored Brands and video campaigns could previously land inside a Sponsored Products row such as Exact or Phrase.
• Fixed Installing the Titan AI plugin in Claude Code from the Titan plugin marketplace no longer fails when you have no GitHub credentials set up. The marketplace now gives Claude Code an https address for the plugin, so the two commands in the Claude Code guide work as written.

• Fixed Keyword rank reads now use the marketplace you ask about. Naming Amazon.ca in a question about your keyword rankings, rank history, tracking limits, tags or segments used to be ignored, and the answer came from your store's home marketplace instead. You can now name Amazon.com, Amazon.de, Amazon.co.uk or Amazon.ca, in chat and in Titan AI MCP. Tracking new keywords still goes to your store's home marketplace.
• Fixed An older chat that was never tied to a store now lets you pick one. If you have more than one store linked, the selector in that chat stays a control, and the store you pick before your next message becomes the one the chat keeps.

• Fixed A question Titan AI asks you now survives a reload. If it stopped to ask you something and you refreshed the page, or opened the chat in another tab or on another device before answering, the question was lost and the chat only looked busy. The question now comes back, so you can still answer it.
• Fixed An answer that is only a table no longer comes with "I didn't produce a response. Want me to try again?" underneath it. The table was there, but because the reply had no written text alongside it, Titan AI offered to run the whole request again.
• Fixed The usage meter in Settings, on the Usage tab, now says At limit when your allowance is used up. A figure that rounded up to 100% used to look exactly like a used-up allowance, and an allowance of zero showed 0%. If you have gone past your allowance, it shows by how much, for example At limit · 185%.
• Fixed When you preview a CSV that Titan AI has attached for you, the preview now tells you when it is showing only part of the file: "Showing the first 200 rows. Download the file for the full data." It has always shown just the first 200 rows, but it never said so, so a 19,000-row report looked like a complete 200-row one. If you were checking whether a filter had worked, there was nothing on screen to suggest you were seeing a fraction of the result. The full file was always in the download, and still is.
• Fixed Replacing the secret on a custom connector now re-tests the connection. Saving a new secret used to update the stored credential and nothing else, so the connector kept whatever status the old key had earned: you could paste a key that no longer worked and the row would still read Connected. Saving now runs the same check as Test connection, so you see the result while you are still looking at the connector.
• Fixed In Titan AI MCP, switching to the account that is already active no longer clears the store you had selected.
• Changed The Titan AI skill and Claude Code plugin are now called titan-ai instead of titan-connect. If you installed the plugin from the Titan plugin marketplace, the old name no longer resolves, so reinstall it under the new name with the two commands in the Claude Code guide. If you download the skill again for Claude or Cowork, remove the old titan-connect skill after uploading the new one so the two do not compete. The old download link now takes you to the new file.

• The auto-approve confirmation can be clicked again. Turning auto-approve on opens a confirmation, and that box was being covered by the chat header, so the buttons did nothing. It now opens above everything else, keyboard focus stays inside it, and Escape closes it.
• If you are at the connector limit, the Add panel now says so before you start. It opens disabled and tells you the limit, instead of letting you fill in the form, run a test connection against your own server, and only then refusing to add it.
• Preview now appears only on files it can actually open. Files between 2 MB and 10 MB uploaded and downloaded fine but showed a Preview button that always failed with "the preview could not be loaded", which read like a broken file. Larger files now offer Download on its own, and when a file is past the preview limit the panel names the file size and the limit instead of reporting a failure.
• The reconnect warning on a custom connector stays readable. When a provider issues no refresh token, meaning you have to reconnect it periodically, that notice used to appear once when you connected and could never be seen again.
• The skill page now shows the date the skill last changed, rather than the date of the most recent deploy.

• Deleting a conversation now asks you to confirm first. Deleting is permanent and removes the files you uploaded to that conversation, so the confirmation names the chat you are about to delete. Previously a single click deleted it straight away with nothing to stop you.
• Branch arrows are back when a message has more than one version. If you edited a message or retried an answer, you can move between the versions again instead of only seeing the latest one.
• The text cursor now sits where you are actually typing when a skill shortcut is in the message box. It used to sit to the left of the real insertion point, so what you typed appeared somewhere other than where the cursor was.
• An approval that cannot be granted now says why. Approve buttons that were quietly dead, with no reason shown, now explain what is blocking them.
• Stores that share a name across marketplaces now show their marketplace everywhere Titan AI lists your stores or asks you to pick one, so a UK and a US store with the same name are easy to tell apart. Stores with unique names are unchanged.

• Fixed When you ask Titan AI which search terms to negate, it reads your whole search-term report page by page before deciding, so a low-spend waste term near the bottom is never missed. On any report longer than about three pages that read used to stop part-way with a "too many requests" message, and Titan AI returned no recommendations at all. Retrying ran into the same wall. Titan AI now paces every request it makes to your advertising data, so long reports are read to the end. If a report is far too large to finish in one go, Titan AI stops after the first page and tells you, instead of working for a minute and then failing in the middle, and suggests narrowing the request to a few campaigns at a time
• Fixed An HTML artifact you download now opens in the light or dark theme your device is set to. It used to always open light, so anything Titan AI had styled for a dark background could be hard to read
• Fixed The arrows for moving between different versions of a message no longer disappear when Titan AI is briefly busy. A rate-limited check used to look exactly like "this conversation has no other versions"
• Improved If you are at the limit of 12 connectors, adding another now tells you so and says to remove one you no longer use. It used to be accepted, but Titan AI could only ever reach the first 12
• New Guides for the Slack and Composio connectors, covering how to connect each one and what it can do once connected

• Improved When the assistant asks you clarifying questions, the question box no longer hides your most recent messages, so you can see the context while you answer. After you answer, the question and your response now stay in the conversation history instead of disappearing
• Improved When the assistant creates a Markdown file, opening it in the preview panel now shows it formatted, with headings, bold text, lists, and tables, plus a toggle to view the raw source. Previously the preview showed the raw markup
• Improved Text and data files (TXT, CSV, TSV, MD, JSON, JSONL, XML, YAML, YML, LOG) now attach at up to 10 MB, raised from 2 MB. Everything else we accept was already at 10 MB or above, so this was the tightest limit on the list and the one most likely to stop a report

The case that prompted the upload change: a member asked for a 60-day search-term analysis, was told to download the report and attach it, and the file came back too large to attach. The report Titan Tools produced could not be given back to Titan AI. That is fixed at both ends. The limit now matches the spreadsheet limit, and for a question of that shape Titan AI works from the data directly rather than sending you out to fetch a file.

This supersedes one row of the table in the 31 August entry. That entry said text and CSV were limited to 2 MB, which was true when it was written. The current limits are always the ones in the chat guide.

• Improved When you add a skill, whether by uploading a file, creating one by hand, or having Titan AI create or update one for you in chat, every check that finds a problem now holds the save, not only the most serious ones. The findings are grouped into a few plain-language rows, worst first, each saying what would go wrong if it were left as it is. This replaces what the 1 September entry said about smaller issues riding along as advisories
• Improved The way forward is Fix with Titan AI. It opens a chat with the skill file attached (the SKILL.md from your upload) and the findings already written into the message, and sends nothing until you do
• New A file that passes its checks shows what it contains before you add it: the files inside with their sizes and the skill's own SKILL.md. The description is read from that file, so you only fill in the name and which stores it applies to
• Note Skills already in your list keep working exactly as they are. The checks run when a skill is saved, never on one that is already there. When you change an existing skill, the parts you change are checked the same way
• Fixed When an uploaded skill file comes back with a long list of checks, the list no longer shrinks to fit the Add a Skill popup and cuts itself off mid-sentence. It scrolls on its own, and the name, stores and Add Skill button underneath it stay within reach
• New The docs now have a Usage and limits guide. It covers what counts as usage and what does not, the allowance included with your plan, the date it resets, and how extra usage works, including that extra usage costs the same rate as your plan with no markup
• Improved When you reach your monthly limit, the notice in chat links to the guide, and so does the Usage tab in Settings, so the explanation is one click from where you meet the limit
• Improved The troubleshooting guide has a new section, "You have reached your usage limit", with what each message means and what to do next

• Improved Uploading a skill file now runs a quick check first. If the file has a problem that would let it install but never actually run, such as no description for the assistant to match on, or a name longer than the assistant reads, Titan AI stops and tells you exactly what to change instead of adding a skill that quietly never fires
• Improved When a check stops an upload, you can send the file straight to a chat with the findings already attached, and fix it there without picking the file again
• Fixed Product image work now stays tied to your real reference photos. If an upload, media ID, or reference-generation step cannot complete, Titan AI does not spend a generation credit on an image it already knows will not match your product. You still receive the completed image brief and prompt, along with the exact reference blocker and the next step needed to generate the faithful version. Generic concept images remain available when you explicitly ask for one after the limitation is explained, and requests for backgrounds, scenes, icons and other non-product creative continue to generate normally without a product reference
• Fixed Connected image-generation services no longer disappear from the assistant's view when an earlier connector advertises a large catalog of tools. Titan AI now preserves a compact routing index for the full connector list while keeping the detailed catalog available separately

Superseded on 2 September. This entry originally said that smaller issues were shown as advisories rather than blocking the upload. That stopped being true the next day: every check that finds a problem now holds the save. See the 2 September entry for what the checks do today.

• Fixed Photos were capped at 5 MB, which is smaller than a lot of modern phone cameras produce. That made images the most restricted thing you could attach, below documents and well below video, and there was no good reason for it. The limit is now 10 MB. If you attached a photo and were told it was too large, it will very likely go through now without you compressing it anywhere else first
• Fixed HEIC photos are now declined when you pick them. HEIC is the format iPhones save in by default unless you have changed it. Titan could never actually read one, so attaching it used to appear to work and then quietly do nothing. Now the composer tells you at the point you pick the file, and tells you what to do instead: on iPhone, Settings > Camera > Formats > Most Compatible saves photos as JPEG, or you can export a single photo as JPEG and attach that. Photos you attached before today are unaffected and still open normally
• Fixed When you asked a follow-up that did not change anything about your request, the assistant could return a different answer than the one it just gave. It now holds the same recommendation unless a real input changed, while still re-reading your live data each time
• Improved Duplicate-keyword checks now apply your definition of a duplicate: the same keyword, at the same match type, for the same advertised product family. The same keyword run across different match types, on a different product family, or used as a negative is no longer flagged as a duplicate, so the checks stop calling your intentional match-type coverage and your negatives a problem. Skills installed before today pick this rule up when you remove it and add it again from Browse Skills
• Fixed On a first sign-in, a new-feature announcement could open on top of the welcome tour, and closing it meant that announcement never came back. The tour now runs on its own. Any new-feature announcement waits until you have finished or closed the tour, and then shows on your next visit with nothing skipped
• Improved When you upload a text or data file (.txt, .csv, .pdf, .xlsx, and similar), the assistant is now instructed to read it with the shell, using the reader that fits the file type. Plain text and CSVs use standard shell tools, PDFs and spreadsheets use the matching converter. It no longer tries to open these as images, which heads off a class of upload read failures

The limits per file type:

We also want to correct something. A changelog entry on 14 July said "file attachments up to 25MB now upload correctly". That change was real, but it raised a ceiling on the request as a whole rather than the per-type limits above, so for photos the number you could actually attach stayed at 5 MB. If you read that entry and then hit a limit, the entry was wrong and this table is right.

The text and CSV row was raised to 10 MB on 4 September. The current limits are always the ones in the chat guide.

• New Not every message needs the same amount of thinking. A level picker now sits beside Send with three settings. Light keeps the reasoning short, which suits casual back-and-forth, a definition, or a small follow-up. Deep works the problem all the way through, which suits a full account review or a campaign restructure. Balanced is the default and behaves exactly as Titan AI did before, so leaving it alone changes nothing. The levels change how much thinking Titan AI does, not how fast it replies, and your choice is saved to your account rather than to one browser
• New Some chats you open every week: a PPC audit, a restock plan, the dashboard you check every Monday. Open the three-dot menu on any conversation in the sidebar and choose Pin, and it moves into a Pinned section that sits above Recent. Pin as many as you want; the most recently pinned sits at the top, and a small pin marker shows which chats you have kept. The same menu reads Unpin, and the conversation goes straight back to Recent in its usual place, and nothing is deleted either way. Collapse the Pinned section by clicking its header, and open it again the same way. Your pins are saved to your account, not to one browser
• Fixed Four PPC reads still answered from your primary store no matter which marketplace you asked about: your negative keywords, your live negative keywords, your live portfolios and your live Sponsored Brands ad creative. A question about Amazon.ca came back with Amazon.com rows, or empty, with nothing saying the store had been swapped. Titan AI now reads all four from the store you selected, and clearly reports when that store is not connected instead of quietly answering about a different one
• Fixed PPC details for Amazon.ca could appear empty or come from the primary Amazon.com store. Titan AI now reads campaign, ad-group, target, product-ad and keyword details from the selected Canadian store. Bid details are included where Amazon provides them
• Fixed When you use Titan through a connector (the TitanConnect MCP server, for Claude and other MCP clients), setting your active seller to a specific marketplace such as Amazon.ca left your PPC reads answering from your primary storefront. The marketplace you set is now pinned, and those reads come back from it until you change it. Asking for a different marketplace on a single question still overrides the pin for that question
• Fixed Listing title reports could present a phrase as one of your supplied keywords even when that exact phrase was not in your keyword list. Reports now verify the whole phrase against your supplied list and keep descriptive, SQP, competitor, and inferred wording clearly separate
• Fixed Approval settings you save on a conversation now apply only for the account that owns that conversation. When using Titan AI through an API key that is not tied to a specific account, a saved approval preference could be read from a conversation belonging to a different account. Those preferences are now always scoped to the owning account
• Fixed A page Titan AI could not open used to come back as a technical error code, which told you nothing you could act on. It now comes back in plain language, and it says which of the two things happened: your list did not cover the site, or the site itself was not responding
• Improved Most sites send you from example.com to www.example.com the moment you arrive. An entry for the plain name does not cover that second address, so a site you thought you had added could still fail to open. Titan AI now tells you when that is what happened. **.example.com is the entry that covers the site and everything under it. When a site is simply down, Titan AI says so and tells you there is nothing to change on your end
• Improved A chart, table or callout in a reply no longer carries a band of empty space above and below it. A reply with several of them is noticeably shorter, so you scroll less to read the whole answer. Every artifact is the same size whether or not it has buttons on it, and nothing moves as your pointer crosses the reply. The button that opens an artifact beside the conversation now sits in the corner of the card, alongside a table's Export button rather than over it
• Fixed Buttons no longer flash gold when you hover them. Secondary controls across the app, including the toolbar above an open artifact, now lift to a soft neutral tint instead. Gold is still the color of primary actions and of the outline around whatever you have selected, so it means something again. Buttons inside an interactive artifact hover in a neutral tone too, so a chart or dashboard you generated reads as your own content rather than picking up the app's colors

• Improved When Titan AI needs a few choices before it can continue, the question card now stays above the message box instead of appearing inside the unfinished reply. Choosing an option moves directly to the next question. Every question includes a custom-answer field, which changes Skip to Next while you type. Use the arrows to revisit questions, choose Skip when you do not want to answer one, or choose Done to finish an unanswered or custom final question. Choosing a final option finishes the card automatically. A single unanswered question keeps the clearer Skip action. Closing the card uses the answers currently shown.

• Fixed A cost-of-goods write where every item fails is now reported as failed. When you set COGS on a batch of products and none of them succeed, the assistant used to be told the write went through, so it could tell you it was done when it was not. That false success is now closed, in chat wherever you use Titan AI and through the Titan AI MCP (TitanConnect). Where the outcome of an item is genuinely unknown, because the write was sent but no result came back for it, the assistant is now told that too and will say the outcome is unconfirmed rather than telling you nothing changed. A batch where only some items fail is unchanged, and reads exactly as it did before

• New Opening a dashboard or report no longer covers the conversation. The chat makes room and stays fully readable beside it, so you can keep talking while you look at the numbers
• New Drag the left edge of an open artifact to balance the split. Wide tables get the room they need, and dragging the other way gives the conversation back. The panel keeps the width you set for the rest of your visit, so you only have to place it once. You can resize with the keyboard too: focus the edge and use the arrow keys, hold shift for bigger steps, or press Home and End to jump straight to the narrowest or widest
• Improved A clearer edge between the artifact and the conversation, so the two read as separate spaces rather than one long column. The maximize button still gives you the full window in one click, and coming back from maximize returns to the width you chose rather than the default
• Fixed Custom reports, the ones you generate from your saved Titan Tools exports, can now come back with their numbers in them when you ask, in Titan AI or through the connector. Asking for one used to hand back a download link and nothing else, and the assistant would then ask you to download that file and upload it again before it could read a single row. Reports whose file is a single table now come back with their rows attached, a page at a time when they are long, so the assistant answers from the figures straight away. Not every report can be read this way, and when one cannot you get the download link on its own rather than a guess at its contents. That covers anything you ask for as an Excel file, the PPC audit, which is always an Excel file, and any report whose file holds more than one table, which includes the dashboard metrics and profit and loss reports. The download link is always there whenever you want the file itself
• Improved Composio connects by signing in. Open Connectors, find the Composio tile and press Connect, and the form is filled in for you: there is no API key to paste and no server address to look up. Press Add, then open the connector and connect it, which sends you to Composio to sign in and approve Titan AI. It reads Needs connecting until you have done that. Composio is a third-party service, not a Titan product: anything it runs is billed there, not against Titan AI, and a connector call that writes, spends credit or changes something asks you first, one call at a time. If you set a Composio connector up the old way, with a key and a server address, it carries on working and there is nothing to redo

• Fixed The community conversations knowledge source is an archive of past group discussions, built from a periodic export rather than a live feed, but Titan AI described it as current and reached for it to answer what is happening now. It now describes the source accurately and routes questions about current activity to the community feed and lessons instead. Community conversations remain the place to look for candid peer opinion and precedent.

• New Web access, in Settings under Web Access. Choose how far Titan AI can reach: off, restricted to a list of sites you approve, or full access. Restricted starts with 28 sites covering the Amazon storefronts you sell in and the research sites members use most, and you can add your own
• New Titan AI can open a page on your list and bring back a screenshot, so it can look at a live listing or dashboard with you instead of asking you to paste it in, and it can click and type on the page too, not just look at it
• New You control the list yourself. Titan AI refuses entries that are unsafe to allow, such as a wildcard over a whole domain suffix or an IP address, and it tells you which rule it hit at the time
• Improved The skills that assumed they couldn't open a link now check your web-access setting first, so they can use it when it's on, though a screenshot is still the reliable path for a gallery or reviews
• Fixed A correction could disappear from the assistant's working context when an existing conversation recovered onto a fresh model thread. Recent conversation context now travels with the recovery request, so the assistant can continue from the corrected guidance
• Improved The assistant now separates Titan AI and project instructions from the files, transcripts and pasted material you supply, treating what you supply as evidence rather than as instructions to follow. Your most recent correction takes precedence over earlier wording, previous answers and the assistant's own inference, and retrieved lessons and tool results now support only what they actually returned. It states when relevant guidance was missing, unreadable or truncated, and surfaces conflicts instead of claiming every source was followed

• Fixed Your connected stores now show you which one actually needs reconnecting, instead of every store reading as connected whatever its real state. You are also only asked to reconnect when the connection has genuinely stopped working, rather than after a brief interruption that clears on its own
• Fixed The billing reset date in Settings is now written the same way for everyone. It used to follow your browser's language setting, so the same date could appear as "August 5, 2026" on one machine and "5 August 2026" on another
• Fixed A conversation you come back to after a long gap now carries on instead of showing an error
• Improved Files you attach now sit inside the message box alongside your prompt, rather than above it
• Improved The top up amount for extra usage is easier to read and enter. Long amounts are no longer cut off, pressing Enter submits the form, and the field settles on the exact amount that will be charged
• Improved Repeated failed sign in attempts on an account are now slowed down

• Improved The account switcher tinted each row amber as you moved over it, which did not match anything else in the app. Hovering a row now shows a subtle neutral highlight instead

• Fixed Reports built fresh from Amazon, the ones you generate on demand rather than your saved Titan Tools exports, now come back with their numbers in them when you ask through the connector. Asking for one used to hand back a download link and nothing else, and the assistant would then ask you to download that file and upload it again before it could read a single row. On a reconciliation spanning months of shipments that is the same errand over and over. Those reports now come back with their rows attached, a page at a time when they are long, so the assistant answers from the figures straight away. The download link is still there whenever you want the file itself, and it still expires 24 hours after the report is generated
• Fixed When you have more than one connected store and it is not clear which one a question is about, Titan AI now asks you to pick instead of answering from a different store. Only accounts with multiple stores were affected
• Fixed New stores now see setup progress while their data finishes importing

• New Composio is in the connector catalogue, one connector covering 250+ tools including Gmail, Sheets and Shopify. Open Connectors, find the Composio tile and press Connect, and the name and the authentication are set up for you. The server address is the one thing only you have: it carries your own server and user ids, so that field starts empty with the shape it expects shown in grey, and you paste the address from your own Composio workspace into it. Your Composio API key is the credential, and Test connection has to pass before Add is available, so a wrong address or key is caught there rather than in the middle of a task. Composio is a third-party service, not a Titan product: the key is yours and anything it runs is billed there, not against Titan AI. A connector call that writes, spends credit or changes something asks you first, one call at a time
• Fixed A connector you just added now reports what it actually is. Any connector added through the form used to land on the list reading Not verified with 0 tools, even though it was connected and working, because the check that ran while you were adding it was thrown away the moment the connector was saved. That check is now kept, so the row shows a verified state straight after you add it, and the real tool count is on the connector's own window, which opens when you press its row. Composio and Higgsfield rows also carry their own logo instead of the generic plug icon
• Fixed Questions about A+ and Premium A+ modules are looked up, not recalled. Asked about module types, hotspots, image sizes or what a Premium module needs, Titan AI would answer from memory, and Amazon changes those rules without announcing it. Titan AI now looks the answer up before it says anything about how the modules work, and for exact specifications and per-module asset requirements it sends you to the upload window in Seller Central, which is the only place those are guaranteed current
• Fixed The format you ask for holds for the whole answer. If you said headline plus three bullets, or set a word cap or a tone, Titan AI would agree and then drift back to its own shape on the following turn. Drafting work passed to a background worker was the main cause: the worker started fresh and never heard your rule, so a skill with its own house format won instead. Your format is now carried into that work word for word and outranks whatever shape a skill would use by default. Titan AI packs more into the shape you set rather than growing past it. To hold a format across a whole conversation rather than restating it, put it in Brand Context
• Fixed Two stores with the same name are told apart in the store picker. If you hold the same business name on more than one marketplace, the in-chat picker listed both rows identically with nothing to choose between them, while every other screen showed the marketplace correctly. The picker now reads the marketplace from the account itself, so the rows read as Amazon.com and Amazon.co.uk whenever two stores share a name
• Fixed Titan AI asks which store instead of picking one. On an account holding several stores, a data question that did not resolve to a specific store could be answered from whichever store came first, and nothing in the reply said which one that was, so the answer read as normal. Titan AI now stops and asks you to choose the store before it reads or marks anything

• New Custom connectors are open. Press Add Connector, then Add Custom, to point Titan AI at an MCP server you run. See Add a connector for the two setup paths and what each one asks for. A connector call that writes, spends credit or changes something asks you first, one call at a time; reading which tools a connector offers does not ask
• New Higgsfield is in the connector catalogue, for generating images and video from your briefs. You sign in with your own Higgsfield account and generations are billed there, not against Titan AI
• New Images and video a connector returns now appear inline in the reply, video with normal playback controls, rather than being cut off with the rest of a long result. Download anything you want to keep: a generated image or video is hosted outside the conversation and its link can expire, so reopening an old chat is not a reliable way to get it back
• New Titan AI documentation is live, covering everything the product does: scheduled tasks, skills, connectors, the chat surface, brand kit and memory, plus setup guides for Claude, Claude Code, Cowork and other MCP clients. The tool reference is generated from the same live registry the product runs on, so what you read matches what Titan AI can actually do, and there is now a changelog with dated entries, newest first, so you can always see what shipped between announcements. Titan AI links to the docs from inside the app
• Fixed Margins are reported, not judged. Titan AI no longer calls a contribution margin thin or healthy unless it can cite a Titan benchmark for that claim, it gives you the number instead

• New Set your cost of goods, and margins go live everywhere. Margin numbers are only as real as the costs behind them, and until now there was no way to hand those costs to Titan AI. You can now set your COGS directly in conversation, up to 500 SKUs in a single approved action, and from the moment they are saved your CM1, CM2 and CM3 margins everywhere are computed from what you set. The approval card shows every value before anything is written, and because all the margin math downstream reads from these numbers, give that card a proper read before you approve it
• Fixed A file you are promised is a file you can click. One path was still slipping past file recovery: when a reply finished in the background, Titan AI could name a file in the text with no download card attached, a promise with nothing to click. That path is now closed as well, the file is recovered from wherever it was written and attached whichever way the reply ends
• Fixed Long messages in the composer type cleanly again. Past roughly ten lines the composer could stop cooperating: the text painted on screen drifted from the text that was really there, and selecting or editing no longer landed where you aimed. What you see and what is actually there now stay in exact agreement however long the message runs
• Improved Slack DM setup catches a wrong ID the moment you save it. The setup expects your Slack user ID, and a channel ID pasted in its place used to save without complaint and then leave DMs looking permanently disabled, with nothing pointing at why. The form now rejects the mix-up as you save and names exactly what to fix, so if you connected Slack DMs at some point and they never came alive, run that step once more and it will tell you what was wrong

• New Billing and extra usage. Your plan is now visible in Settings. Until now, reaching your plan's included usage meant waiting for your cycle to reset. You can now switch on extra usage in Settings and Titan AI keeps working past the limit, drawing on a balance you load yourself, any amount from $1, as a one-time top-up rather than a subscription. Your included allowance always spends first, the balance covers only what runs beyond it, and what you load never expires. The balance stays visible in Settings, and the top-up controls appear once extra usage is switched on
• Improved Approval cards show every argument. When Titan AI asks you to approve an action, the details used to be squashed onto one line and cut short, so a long field could push an important value out of sight. Each value now gets its own labelled line, so the thing you are deciding on is fully in front of you before you approve it
• Fixed Reports with no data now tell you so. When you asked for a report over a date range with no activity, Amazon returned a file containing only the column headings and Titan AI handed you the download link anyway, leaving you to open it to find out it was empty. Titan AI now checks the file before offering it and tells you there is no data for that range instead
• Fixed File paths in replies read correctly. When a reply mentioned an internal file path, Titan AI replaced it with a note saying the file had been removed, even when the download was sitting right there in the same message. That note now says only that the path is hidden, and no longer claims anything was removed

• Fixed Approval stays required on tasks that run while you are away. A scheduled task could reach the point of offering to send a Slack message on a turn where no approval card is ever put in front of you, which meant a message could go out from your workspace without you having seen it and agreed to it first. That action is no longer offered on scheduled task turns at all, so the rule you rely on, that nothing gets sent in your name until you have approved it, holds on the runs that happen without you watching as well as the ones you are sitting in front of

• New Titan AI can now pull seven further live views of your advertising account from inside a conversation, including the ones its PPC skills lean on most heavily, along with three more capabilities available for the first time. In practice that means fewer moments where Titan AI tells you it cannot see something on your account that Amazon reports plainly
• New Files arrive in Titan Tools Copilot too. When Titan AI builds a file for you inside the Titan Tools embed, it now arrives as a download card in the conversation exactly as it does on the web app, closing the last surface where a file could be named in a reply and never actually reach you
• New A weekly PPC cadence skill is now in the official catalog. Run the full weekly PPC routine against your own account with one command, rather than rebuilding the cadence yourself every week
• New Your feedback now has a place to live, and a status you can follow. When you report something to Titan AI it becomes a tracked item with a reference and a visible status, so instead of a message that disappears you can come back and see exactly where your report stands. Live today at ai.titannetwork.com/feedback
• Improved Brand Kit editor UI rebuilt. The editor now opens as a split dialog that shows your brand as you build it, the font library is considerably larger than the set it launched with, and both the font and color pickers take typing as well as clicking, with the font list scrolling properly under a wheel or a trackpad. It is on Titan AI on the web and in the community chat
• Fixed Negation and search term reads now page through everything. A negation session pages through the full set of search terms rather than stopping at the first page and treating what it had as the whole picture, and where a page genuinely is partial the result now says so rather than presenting itself as complete
• Improved Approvals behave whichever way they end. Members part way through setup could not approve a skill or task card at all, and the request behind it would sit until it timed out and then report itself as denied, which made it look as though you had refused something you were never actually shown. Approvals now work regardless of where you are in setup, and an approval that does expire tells you plainly that it expired and what to do next
• Improved Placement data is applied. When a placement request names specific campaigns, those campaigns are now honoured instead of being ignored while the result still came back looking complete, and changes to placement modifiers are read back and confirmed after they are written, so a placement change you asked for is one you can trust landed
• Improved Usage counting stays on your side. Resuming an older conversation no longer risks counting the same usage against you twice, and Titan AI now keeps working rather than blocking you if usage configuration is briefly unreadable
• Fixed The monthly MCP request meter no longer counts your chat. That meter was only ever counting conversations inside the app rather than the Titan AI Connect requests its name described, so it has been removed along with the cap that sat on it. Nothing you do in chat counts against it and it no longer appears on your Usage tab
• Fixed Shared conversation links show a neutral title. The title Titan AI writes for a conversation can pick up your store or brand name, and it was appearing in the preview card that renders when a share link is pasted somewhere public, as well as in the page's own public metadata. Those pages now carry a plain Titan AI title in both places, and they are marked so search engines leave them alone. The change applies the moment a page is requested, so every link you have already shared is covered with nothing for you to do, and what someone sees after opening a link you gave them is unchanged
• Fixed Requests that stall upstream at Amazon are now identified as timeouts and given a ceiling rather than hanging, and a store that is temporarily unreachable no longer reports itself as you having no linked account at all

• Fixed Negative keyword reads now state their own scope. A negative keyword read on a campaign reaches the negatives held at ad group level, and campaign level negatives sit outside what it returns, so Titan AI now says exactly that whenever it makes that read rather than letting an empty result stand on its own. It will also no longer propose rebuilding a negative list off an empty return. If an audit or a negation session you ran recently came back showing few or no negatives on a campaign, it is worth confirming your campaign level negatives directly before you act on it
• Fixed The store chip always names the store your conversation is running on. Every reply now carries the store identity resolved for that turn, so it stays accurate through sending, through the move into a new conversation, and through a reload. On an account with more than one linked store, a new chat now opens with no store selected and waits for you to choose, rather than picking one for you. Your messages have always run against the store you selected, so there is nothing to revisit on work you have already done
• Improved The conversation stays live while an artifact is open. Opening a document, a chart, or a delivered file no longer holds the conversation behind it, so you can scroll it, click into it, and keep typing with the artifact open beside you. The panel toolbar is also tidier, with a single copy control and download now one click rather than two
• Improved Artifacts stay readable in whichever theme you use. The agent now builds artifacts knowing the frame follows your light or dark setting and changes with it, and it measures the text it is given against the surface it is painted on and lifts anything sitting below a legibility floor. That applies to artifacts you already have as well as new ones, so older ones should read better from now on, and regenerating any that still look off will produce a cleaner result
• Fixed Direct messages to Titan AI in Slack now work. To use it, connect your workspace under Settings, then Slack, in the web app

• Fixed File delivery. When the agent names a file in its reply and then fails to attach it, that file is now recovered from wherever it was written and attached anyway, so a filename mentioned in the text becomes a download card in the chat rather than a promise you cannot act on
• Fixed PPC audits. Five official PPC skills could return inaccurate results: ppc-starved-targets-audit and ppc-waterfall-check could fail on their primary path, and placement metrics came back account-wide on every campaign rather than per campaign, which ppc-waterfall-check and ppc-alexa-optimizer build their placement attribution on. Campaign and ad group filtering now works end to end and the placement split is genuinely per campaign, so if you ran any of those three recently, it is worth re-running it now
• Improved Actions approvals. When the agent proposes a change on your account, the approval card is now bound to the specific action that raised it, so a background worker can no longer carry out a write that never appeared on a card in front of you. The card describes only what the message actually shows, and when the agent is simulating rather than executing, it now reads as a dry run instead of looking like a change that already landed. If you have explicitly blocked a tool, that block now outranks auto approve and anything agreed earlier in the conversation
• Fixed Shared conversation links. A share link could reveal which Amazon store a conversation was about, both in the agent's visible thinking text and in data embedded in the page itself, to anyone holding the link. Both paths are now closed, and every link you had already shared was clean the moment this deployed, with nothing for you to do. Shared transcripts now show the agent's written answers only, not its thinking, so a turn where the agent reasoned and used tools without writing a closing answer can look like a question with no reply. If you would rather a link stopped resolving altogether, Share Conversation has a "Revoke Share Link" button that kills it immediately
• New Images and files in Titan Tools. Attachments you upload inside the Titan Tools embeds now still render after you reload the conversation
• Fixed Pasting or dragging a file type the composer does not support now names the file it is rejecting instead of failing at the end of an upload, and a file you add while another one is still uploading is queued rather than silently dropped
• Fixed Email capitalisation. Members whose address was stored with different capitalisation from one app to the next could find their files unreachable across apps, and in some cases a Slack workspace showing as connected in Settings while the agent reported no connection at all. Addresses are now stored one canonical way everywhere, existing accounts included, so your skills, Brand Kit, brand context, scheduled tasks, and Slack connection follow you across every Titan AI surface
• Improved Brand Kit editor. A full design pass so every field is laid out the same way, the guidance sits in one place instead of drifting between the title and the card, the information tooltips are legible in dark mode, and the "Applies to" store selector now reads as part of the control it belongs to

• New Brand Kit. In Settings you can now save your visual identity, your logo, brand colors, and fonts, and the agent keeps it on hand while it builds for you, so decks, dashboards, and creatives come back carrying your brand instead of a generic look. Images are coming soon
• New Real PDFs and file bundles. The agent now renders proper PDF documents when you ask for one, and bigger jobs can arrive as a single .zip or .tar.gz bundle instead of a stack of separate downloads
• New Skills carry their files. A skill you create in a conversation can now package the files from that conversation with it, so the template or checklist you refined together travels with the skill. Invoking a skill with / now keeps it active for the whole conversation instead of dropping it after one reply, so a long session follows your playbook start to finish
• New A cleaner home for artifacts and files. Documents the agent builds now open in their own side panel with preview, code view, and download in one place, delivered files use the same panel, and charts and tables stay right in the conversation. Chart requests are also more reliable on the first ask
• Improved File delivery now runs on a single confirmed path: the agent receives confirmation that your download card exists before it tells you the file is ready, so a promised file is one you can actually download
• New Feedback with Loom video. The feedback form in all three Titan AI apps now takes an optional video link, so you can drop a Loom of what you are seeing
• Fixed Attaching a file type the composer does not support now gets you a clear message on the spot, instead of the picker closing and nothing happening
• Fixed Images and files you upload in chat now survive a conversation reload instead of going blank
• Fixed Connecting a workspace that is already connected now tells you so, and by whom, instead of closing the window without a word
• Fixed Claude Code sign-in now completes whichever port your CLI picks, and the consent screen states exactly what you are authorizing, with your account and skills shown correctly
• Improved Search Query Performance pulls are no longer limited to 20 ASINs per call
• Fixed The Scheduled Tasks tooltip is readable in light mode again
• Improved The welcome email new members receive now describes exactly how setup flows

• Improved Titan AI's primary address is now ai.titannetwork.com. Older addresses (asktitan.titannetwork.com, titanai.titannetwork.com, and asktitan.com) still work and redirect you there automatically
• Improved The first time you open the new address, you will need to log in again

• Fixed Campaign budget changes go through more reliably. Where a change is missing a detail Amazon requires, Titan AI names the detail to add rather than failing on it. See Actions on your account for how a change is proposed and approved
• Improved When your Amazon connection needs refreshing, Titan AI points you at the place to reconnect it instead of leaving you to find it

• New More file types in chat. Alongside the spreadsheets, CSVs, PDFs and images Titan AI already reads, you can now upload JSON and JSONL, XML, YAML, TSV and log files, plus Word and PowerPoint documents. In practice that means the files your Amazon business already runs on, an API export of your keywords, a bulk operations sheet, a listing flat file, your brand guidelines document, go straight into the conversation and the agent works on them directly, and if a data file arrives malformed you can simply ask it to repair the file for you
• New A clearer start for new members. The setup screen right after connecting an account is now driven by the live status of the data import itself, so instead of a generic loading page new members watch their orders, campaigns and keyword history arrive step by step and know exactly when the account is ready to work with. See Your first week for what that setup covers
• Fixed Conversations survive any character. Messages containing unusual characters, the kind that ride along when you paste listing copy, review text or foreign-language keywords, can no longer break a conversation mid-analysis
• Fixed Files from longer tasks always arrive. When the agent runs a longer job for you and builds files along the way, a keyword list, a PPC audit export, a report file, those files are now always delivered into the chat instead of occasionally going missing at the end of the run
• Fixed Uploads inside Titan Tools. Attaching files in the chat embedded in the community and in Titan Tools works properly again, so you can hand the agent a file from wherever you are chatting
• Improved Brand Context everywhere you chat. Brand Context now loads everywhere, including on browsers that block cross-site cookies, Safari for example, so the agent keeps your brand knowledge with it on every surface
• Fixed Your store stays selected. Reopening an older conversation keeps the account that analysis was run on instead of silently switching to your first account, which matters when you manage several accounts and move between past analyses
• Improved Switching to a disconnected account. If you switch to an account whose Amazon connection was revoked, Titan AI now tells you plainly to reconnect it from the dashboard instead of failing with an error
• Fixed The right screen at your usage limit. If you reach your plan's usage limit you now land on your actual usage screen, where you can see exactly where you stand, instead of a guest signup prompt
• Fixed Skills menu and slash commands. The Skills flyout and slash command panel now open exactly where they should and behave correctly throughout the chat box
• Improved Login security. A security update to the login system shipped, nothing changes on your side

• New Live Amazon Reports. You can now ask Titan AI to generate Amazon reports for you on demand, the same reports Seller Central builds, 15 types covering sales and traffic, orders, returns, listings, FBA inventory health, the inventory ledger, fulfillment and storage fees, and reimbursements. Ask in chat wherever you use Titan AI, or through the Titan AI MCP (TitanConnect), and the report is created fresh on your account and comes back as a download link once Amazon finishes preparing it, usually within a few minutes. Each link stays valid for 24 hours, and you can request a fresh one any time. See Live Amazon reports for the full list
• Fixed Usage counting. Plan usage limits were counting some conversations several times over, so recorded usage could run well past real spend, and a few members hit their limit on numbers that were not their actual usage. Counting is now exact per turn
• Fixed Skills in the composer. The Skills menu now opens in its own flyout that stays pinned while you browse, and slash commands autocomplete with full keyboard navigation, arrows to move and Enter to pick, so reaching the right skill is faster and needs no mouse

• New Waiting-on-you marker. The sidebar already shows a running indicator and a green marker on finished replies you have not opened, and it now adds an amber marker on any conversation paused waiting on something from you, an approval for example, so you can spot at a glance where the agent needs you before it can continue
• Improved Conversation titles. New conversations now name themselves seconds after your first message, while the reply is still being written
• Improved Send and step away. Your message is saved the moment you send it, so switching to another conversation mid reply loses nothing, the reply keeps working in the background and is waiting in the thread once it finishes. Pressing stop now closes every section of the answer cleanly instead of leaving spinners running
• Fixed Approval cards. When the agent proposes a change on your account, the approval card now shows up dependably mid reply, including a second proposal in the same answer, so a proposal no longer sits invisible until it expires
• Fixed Chat inside Titan Tools. The embedded chat now survives page navigation, and when you come back it reconnects and resumes your conversation exactly where you left it
• Improved Bigger tasks. Multi-part requests now reliably split across parallel background workers, so a wide job runs side by side instead of one step at a time
• New Account data. The agent now knows the newest inventory numbers on your account, your on-hand and FC transfer quantities, along with the new closing-fee data
• Fixed Titan AI Connect. Year-over-year comparisons now stay accurate when Amazon moves a product under a new parent ASIN between the two periods, and a product missing from a top-10 view means it fell below the top 10, not that it sold nothing, so an identifier change can no longer show up as a false zero in your comparison
• New Skills. The skills list now shows on-request skills as well, each with a note on how it activates, so you can see everything available to you, not only what runs on its own
• Fixed Started as a guest. If you chatted before logging in, those conversations now carry into your account when you log in, whichever way you sign in
• New Usage limits. Your plan's usage limits are now enforced in the app, and everything the agent spends on your behalf, background workers included, counts against the same allowance

• New File downloads. When a skill finishes and tells you your files are ready, they now come through as download cards you can open straight away, wherever the agent happens to save them along the way. A skill run that produces a report, a dashboard or a spreadsheet reaches you every time, and this applies across every skill
• Fixed Moving between answers. When you edit a message and the agent gives you a second version, the arrows that move you between those versions now work across all of your conversations, so you can compare the two and keep the one you want
• Improved Cleaner answers. The text and errors coming back from background workers now stay in plain seller language, keeping system detail out of your chats

• Fixed Broken links. Dead links to a dashboard, report or HTML file the agent produced are gone from every message across all Titan AI apps. Files that arrive keep coming as download cards, and when one doesn't, the message now tells you plainly so you can ask the agent to attach it, instead of pointing you at a link that goes nowhere
• Fixed Accounts management. Starting a chat now respects the Primary account you set and lets you work on any of your linked accounts again, and conversations stay on the marketplace you started them in, so no more flipping from US to UK mid-thread. Store rows carry their marketplace label at chat start as well, and if connecting an account ever fails, the window now reports the actual error instead of hanging silently. See Accounts
• Improved Skills. When you run a skill, its instructions now take priority over the network's general learnings, so the agent follows your playbook start to finish instead of veering off to Titan knowledge partway through. Official skills also declare what they need to run, a skill that depends on something the platform can't do yet, like browsing the outside web, now says so upfront instead of quietly answering from its general knowledge. Skills installed before today pick this up on a reinstall from Browse Skills. See Skills
• Improved Agent language. Titan AI now narrates its progress and writes its answers in plain seller language, no more jargon like scope tables, structural endpoints or join tables. The same discipline applies to errors: when a background worker gets stopped mid-task, the chat now shows what it managed to produce and says so plainly, instead of painting a red error block over your results
• Improved Faster chats. Opening a conversation was firing one request per message, hundreds at once for long chats. It now fires exactly one, so conversations open faster and hold up better during busy hours
• Improved The section introduced at launch as Missions is now called Scheduled Tasks everywhere in the app, the same capability under a name that says what it does. Everything you already set up there is untouched
• Improved Integrations moved to Settings, and now includes a step-by-step tab for connecting Titan AI to Claude or any AI host through the MCP connector

• Fixed A traffic bottleneck that was causing intermittent errors and slow loading in chats, especially during busy hours. The same issue could make some agent tool calls fail mid conversation. Both are resolved
• Fixed The overall size ceiling on an upload was raised to 25MB. Previously, anything over 10MB could fail with an error. Corrected 2026-08-30: this raised the ceiling on the request, not the per-file-type limits, so photos stayed at 5MB until the 30 August entry. See that entry for the limit on each type
• Improved Clearer error messages. If the AI service is ever temporarily overloaded, you'll now see "the service is busy, please try again in a few seconds" instead of a confusing message about plans and billing. Nothing was ever wrong with your Titan AI plan and there's nothing to change on your side
• Improved Reliability of how Titan AI splits big requests into parallel research workers behind the scenes

• Fixed Skills now reach the agent in every conversation, so it follows your playbooks instead of falling back to its general Titan knowledge. The / menu in the chat lists them all. See Skills
• New Session status in the sidebar. Every conversation now shows where it stands without opening it, a running indicator while Titan AI is still working and a green marker on any finished response you haven't looked at yet. Both survive page reloads, so you can start several requests, step away, and see exactly what finished when you come back
• Fixed Stores linked before this update now carry their marketplace label too, in the same "Store Name - Amazon.co.uk" format as newly linked stores
• New Action alerts. When Titan AI proposes an action on your account, bids, negatives, budgets, campaign state, it waits up to nine minutes for your approval. The tab now plays a soft chime the moment a proposal needs you, sends a browser notification if you're away from the tab, and keeps a counter on the tab title until you respond. Accept the notification permission prompt the first time so proposals reach you even when you're working in another tab. See Actions & safety
• New Files from your agent. Everything Titan AI produces for you, exports, reports, spreadsheets, now arrives in the chat as a download card, with previews opening in a side panel. This includes a single Excel workbook with multiple tabs instead of table-by-table downloads. See Chat results

• Fixed Store labels. If you manage stores that share a name across regions, every account you link now carries its marketplace in the label, and accounts linked earlier pick theirs up too. See Accounts
• Fixed Dashboards. Interactive dashboards you ask for now render live in the chat as charts and one-screen views, instead of coming back as pasted text
• Fixed Skills. The workspace where skills run and write files is working again, and the full set of 27 official Titan skills is live in Browse Skills. See Skills

• New A new way of working. Tell Titan AI the outcome you want and it breaks the job into steps, brings in specialist agents where it needs them, and you watch the whole thing happen live in the session
• New Full-catalogue coverage. A PPC review no longer means reading just the first page of your data. The new version splits the job across specialists and works through every campaign, every target and every SKU
• New Action alerts you approve. When Titan AI wants to change something on your account, bids, negatives, budgets, campaign state, it shows you exactly what it intends to do and nothing runs without your approval. For routines you trust, an auto-approve switch in the chat header (off by default) puts the level of autonomy in your hands. See Actions & safety
• New Clean separation between your brands. Titan AI works on the right account's data whether you're chatting, building a skill or running a mission, and you can give each brand standing context that it picks up every time that brand comes up. See Accounts
• New Skills, your playbooks made permanent. Your weekly brand report structure, your rules for negating search terms, your launch checklist, all of it can be saved once as a Skill and run any time by typing / in the chat. A skill can apply to one brand or to all of them, and official Titan skills are built in so you can start from proven ones. See Skills
• New Missions, jobs you hand over. The new Missions section lets you set up the work you repeat, a Monday morning account review, a wasted-spend sweep, an inventory check before reorder, and run it on demand whenever you want it done. Scheduled runs that fire on their own are coming next
• New Connectors, one place for everything Titan AI reaches. The new Connectors page shows what is linked to your Titan AI, starting with Slack: connect your workspace, tag Titan AI in a channel, and ask about your account without leaving the conversation your team is already in. Google Workspace is coming soon. See Connectors
• New Outputs in the shape you need them. Reports come back as files you can download and share, and you can hand Titan AI your own files to work from, PDFs, spreadsheets, CSVs, images, and it works through them the same way it works through your live data
• New Dashboards inside the conversation. Ask for "my ACOS by campaign over the last 30 days as a chart", a keyword table ranked by wasted spend, or a one-screen PPC dashboard, and Titan AI builds the visual right there from your data. See Chat results
• New A sharper overall experience. Answers pull exactly the slice of data you ask for so they come back faster and more precise, conversations survive reloads and back-navigation mid-answer, chats get sensible titles, images open full size on click, and new members get a guided tour on their first visit