What's shipped.

Three plans: Free, Lite at $5 and Pro at $15

• •The plans are now Free, Lite and Pro. Free is unchanged: 20 AI requests a day on Meta Muse Spark 1.3, no card. Lite is $5/month or $48/year and carries 100x more usage than free on Meta Muse Spark 1.3 Contributor with no daily request cap, plus file attachments and screenshots. Pro is $15/month or $144/year and carries 7.5x more usage than lite across every included model: Kimi k3, DeepSeek V4.1 Flash, DeepSeek V4 Pro, Gemini 3.8 Flash, Meta Muse Spark 1.3 standard and Meta Muse Spark 1.3 Contributor. Pro also takes your own API keys and adds the QuickBooks, Campfire and Stripe connectors.
• •Nothing you had to do. The old $5/month plan is Lite, and the old $15/month plan and the $25/month connectors plan merged into Pro. If you were on EBITDAI Max, you are on Pro now, with the same models, the same own keys and the same three connectors. If you were on the $15/month plan, you are on Pro with the same model list and a smaller included allowance than the old plan carried. Open the pane to see your plan; no reinstall, no new sign-in.
• •Lite is a paid plan on Meta's Contributor tier, and it has to be said plainly: Meta trains on every prompt, reply and workbook value you send with Meta Muse Spark 1.3 Contributor, which is the only model Lite runs. Pro can run the Contributor tier too, or the standard Meta tier, which Meta does not train on. Lite turns are not written to the free-plan review log, so our team does not read them.
• •We retired the usage multiples in this change. The old 6x and 20x claims were measured against the old $9.50 monthly allowance, and that allowance is gone, so the claims are gone with it rather than re-measured. The plans now state their included usage in dollars a month, and the homepage chart still shows what those dollars buy in complete model builds.
• •Entries dated before September 13, 2026 keep the name the plan had at the time.

Free plan doubles again to 20 requests a day

• •The free plan is now 20 AI requests a day on Meta Muse Spark 1.3, up from 10, still with no card. It refreshes daily, and the model is unchanged: Meta serves it on Meta's Contributor tier, so Meta trains on every prompt, reply and workbook value you send with it.

Kimini is now DeepKimini: DeepSeek V4.1 Flash is the default model

• •The $15/month plan is now DeepKimini: DeepSeek V4.1 Flash is the default model. Same plan, same $15/month or $144/year ($12/month), same included usage, same own keys; nothing to do for existing subscribers.
• •The name matches what the plan runs: DeepSeek V4.1 Flash, DeepSeek V4 Pro, Kimi k3 and Gemini 3.8 Flash on one shared allowance, both Meta Muse Spark 1.3 tiers, and your own API key for every provider.
• •The landing-page chart now carries two DeepKimini bars: the $15/month allowance buys up to 1,727 complete model builds a month on Meta Muse Spark 1.3, the cheapest model in the plans, and up to 846 on DeepSeek V4.1 Flash, the default. Against ChatGPT Plus at roughly 40 tasks and Claude Pro's best case of 85 builds, that is up to 40x ChatGPT Plus and 20x Claude Pro. The old flat-rate Kimi Code bar, which carried the larger published multiple, is gone with it.
• •Entries dated before September 10, 2026 keep the name the plan had at the time.

DeepSeek Flash upgraded to V4.1 on every plan

• •DeepSeek serves its Flash model as DeepSeek V4.1 Flash now, and every plan that had V4 Flash runs it: DeepMuse, DeepKimini and EBITDAI Max. It is one entry in the model picker at one price, not a second model. The free plan is unchanged and still runs Meta Muse Spark 1.3 only.

Attachments move to paid plans

• •File attachments and screenshots in the add-in chat are now part of every paid plan. On the free plan the attach button still opens, and shows a short note that attachments are included from DeepMuse at $5/month. Nothing changes for DeepMuse, Kimini and EBITDAI Max.

Free plan doubles to 10 requests a day, and the Meta Muse Spark write-up

• •The free plan is now 10 AI requests a day on Meta Muse Spark 1.3, up from 5, still with no card. It refreshes daily. The model is unchanged: Meta serves it on Meta's Contributor tier, so Meta trains on every prompt, reply and workbook value you send with it, and EBITDAI logs free-plan conversations for product review.
• •New post: Meta Muse is now on every EBITDAI plan, the cheapest model we have ever run. What each plan gets, a token price table against DeepSeek, Gemini, Kimi and Claude Sonnet, our measured cost per modeling task on the hard suite, what the included usage buys in model builds, and the Contributor tier training terms behind the price.
• •The homepage and pricing page intros now name all four model families we run: DeepSeek, Kimi, Gemini 3.8 Flash and Meta Muse Spark 1.3.

DeepMuse at $5, EBITDAI Max at $25

• •The DeepSeek plan is now DeepMuse, at $5/month or $48/year ($4/month). It runs DeepSeek V4 Flash, DeepSeek V4 Pro and Meta Muse Spark 1.3, still takes your own DeepSeek key, and still has no connectors.
• •Included usage on DeepMuse is smaller than the old $10 plan carried, so the claim moved with it: DeepMuse matches or exceeds a $20 Claude plan rather than beating it by a multiple. Kimini and EBITDAI Max are unchanged at up to 240% more.
• •EBITDAI Max is now $25/month or $240/year ($20/month). Same models, same included allowance and same own keys as Kimini, plus the QuickBooks, Campfire and Stripe connectors. Layered with a $19 Kimi Code subscription the stack is $44/month.
• •Kimini is unchanged: $15/month or $144/year ($12/month), Kimi k3, DeepSeek V4 Flash, DeepSeek V4 Pro and Gemini 3.8 Flash on one shared allowance, both Meta Muse Spark 1.3 tiers, and your own API key for every provider.
• •The pricing page carries the new prices, plan cards and comparison rows, and shows each paid plan's previous or list price struck through beside the live one.

A Meta tier Meta does not train on, and a training warning you can see

• •Kimini ($15/month), EBITDAI Max ($19.99/month) and the trial can now pick Meta Muse Spark 1.3 on Meta's standard tier, shown in the model picker as "Meta Muse Spark 1.3 (not used for training)". Same model as before, different terms: Meta's published pricing terms say content submitted on that tier is not used to train Meta's models.
• •The Contributor tier stays, shown as "Meta Muse Spark 1.3 (trains on your data)", and it still sends every prompt, reply and workbook value to Meta for training. Neither Meta option is ever the default. The DeepSeek plan ($10/month) and the free plan run the Contributor tier only.
• •The training warning moved out from under the model picker. Pick the Contributor tier and the pane now shows a bar at the bottom spelling out what goes to Meta, with an X to dismiss it for the rest of that session. It comes back the next time you open the pane, and whenever you switch back to that model.
• •The pricing page, the security page, the Privacy Policy and the Terms of Service now say which Meta tier Meta trains on.
• •Superseded September 7, 2026: the DeepSeek plan was renamed DeepMuse and repriced to $5/month later the same day.

Meta Muse Spark 1.3, and a new model on the free plan

• •Meta Muse Spark 1.3 joins the included models. It is selectable on the DeepSeek, Kimini and EBITDAI Max plans and on the trial, and it is never the default: the default model on your plan is unchanged.
• •Meta serves it on Meta's Contributor tier. Every prompt, reply and workbook value you send with that model goes to Meta, and Meta uses it to train its models. The pane says so under the model picker whenever the model is selected. Pick another model if that is a problem for a workbook.
• •The free plan now runs Meta Muse Spark 1.3 only, still 5 requests a day, no card. DeepSeek V4 Flash is no longer a free-plan model. EBITDAI also logs free-plan conversations, as it has since September 2.
• •Superseded September 7, 2026: the Meta warning is now a dismissible bar at the bottom of the pane rather than a note under the picker, and Kimini and EBITDAI Max can also run Meta Muse Spark 1.3 on Meta's standard tier, which Meta does not train on.
• •The pricing page, the security page, the Privacy Policy and the Terms of Service now spell this out.

Own API keys on Kimini

• •Kimini ($15/month) now connects your own API key for every provider: Kimi Code, Moonshot (pay-as-you-go Kimi), DeepSeek and Gemini. Same own-key support EBITDAI Max has, at the Kimini price.
• •On your own key you pay your provider its published prices with zero markup, and usage is whatever your provider allows. That path used to begin at EBITDAI Max; it now begins at Kimini. DeepSeek and pay-as-you-go Kimi calls go browser-direct; Kimi Code and Gemini cannot be called from a browser, so they pass through our stateless relay, which stores no prompt or workbook content.
• •EBITDAI Max ($19.99/month) is now Kimini plus the QuickBooks, Campfire and Stripe connectors. Same models, same included usage, same own keys, plus live financials pulled straight into the workbook.
• •Nothing changes for existing subscribers on either plan, and there is nothing to do. Connect a key from the account panel in the pane.

Free-plan conversations are now logged, and the fine print says so

• •To learn what free users need and what frustrates them, EBITDAI now keeps free-plan conversations (prompts, replies, the analyst's actions and the workbook context sent to the model) for 90 days, and our team reads them. This is the free plan only.
• •Paid plans are unchanged: on DeepSeek, Kimini and EBITDAI Max, prompts and workbook content pass through our relay without being stored, and EBITDAI Max on your own key never touches our servers at all.
• •Superseded September 6, 2026: own API keys start at Kimini, and the never-touches-our-servers part holds for own-key DeepSeek and pay-as-you-go Kimi only. Kimi Code and Gemini cannot be called from a browser, so those calls pass through our stateless relay, which stores no prompt or workbook content.
• •The pricing page, the security page, the Terms of Service and the Privacy Policy now spell this out.

Kimi+ is now Kimini, and it runs Gemini too

• •The $15/month plan has a new name: Kimini, a play on Kimi + Gemini. Same price, same included usage, nothing to do for existing subscribers.
• •Kimini and EBITDAI Max now run Gemini 3.8 Flash (launched today) on included usage alongside Kimi k3 and DeepSeek V4, switchable mid-session.
• •We ran Gemini 3.8 Flash through the published benchmark the same day: 28 of 30 tasks (93%) at about ten cents of model cost per task, and the fastest model on our rig. Full results and the two misses are on the benchmark page.

Connectors on EBITDAI Max, Gemini keys are back, and a faster add-in

• •QuickBooks, Campfire and Stripe connectors are now part of EBITDAI Max: ask the analyst to pull your real P&L, balance sheet, invoices or customers straight into the workbook.
• •Google Gemini joins Kimi and DeepSeek as a key you can bring on EBITDAI Max (Gemini Flash and Gemini Pro, switchable mid-session).
• •Superseded September 6, 2026: own API keys are not an EBITDAI Max exclusive. Kimini ($15/month) connects your own key for every provider too, and the connectors are the only thing EBITDAI Max adds.
• •Your own API keys are now saved to your account, encrypted, so you never re-enter them after a session ends. On Google Sheets they persist too.
• •Usage & limits shows one bar per included model, plus an External APIs section for your own keys with an estimated spend for pay-as-you-go providers.
• •Big edits are much faster: a new fill tool writes whole blocks of formulas in one step, the analyst reads sheets with row numbers so it stops miscounting, and it jumps to the last block it wrote and names every tab it changed.
• •If Kimi's or DeepSeek's servers are overloaded, the pane says so plainly, retries automatically, and offers a one-click switch to the other model. Partial work is kept if a turn fails, and saved chats keep every reply.
• •The chat composer is redesigned: one rounded box with attach, chat history, the model picker and send all in one row.

Private Mode is now EBITDAI Max

• •The top plan has a new name. Private Mode is now EBITDAI Max: same $19.99/month, same $192/year, same features, including running on your own Kimi or DeepSeek key so your workbook data never transits our servers.
• •Superseded September 6, 2026: own API keys now start at Kimini, and only own-key DeepSeek and pay-as-you-go Kimi calls skip our servers. Kimi Code and Gemini calls pass through our stateless relay, which stores nothing.
• •Nothing changes for existing subscribers and there is nothing to do. Your plan, billing date, and included usage are untouched.
• •The pricing page is rebuilt around a side-by-side comparison of all four plans, so you can see exactly what each one includes before you pick.

Free plan: 5 AI requests a day, and a fix for early lockouts

• •The free plan now includes 5 AI requests a day on DeepSeek (up from 3), refreshing daily, no card.
• •Fixed: one large request could use up the whole day's free allowance and then show a 'you've used your 3 free requests' message after a single request. The free plan now runs DeepSeek V4 Flash, the daily allowance is sized for all 5 requests, and the message says what actually happened.
• •Superseded September 6, 2026: the free plan now runs Meta Muse Spark 1.3.

Free plan: 3 AI requests a day, no card

• •EBITDAI now begins for free: every account gets 3 AI requests a day on DeepSeek, refreshing daily, no credit card. Sign in and start modeling.
• •Existing accounts get it automatically -- if you signed up before and hit the free-plan limit, open the add-in again.
• •Paid plans start immediately at checkout, and everything else about them is unchanged.
• •Superseded September 6, 2026: the free plan now runs Meta Muse Spark 1.3.

Kimi+ and bigger included usage on every plan

• •Included monthly usage went up on every plan -- every plan now includes more modeling usage than a $20 Claude plan allows, dedicated to Excel. Existing subscribers get the bigger allowance automatically.
• •The Kimi plan is now Kimi+: usage on both Kimi k3 and DeepSeek, with the ability to reload either anytime. Same $15/month.
• •Reminder: your usage is a monthly pool -- no 5-hour windows, no weekly lockouts.

Click a cell reference to jump to it

• •Any cell or range the agent mentions in chat -- B2:D14, Summary!A1, a range in a tool step -- is now a link. Click it and Excel activates that sheet and selects the range.
• •Works in summaries, streamed replies, and the activity log, on every plan.

Referrals: give a month, get a month

• •Referrals are live: every active subscriber gets an unlimited-use referral code (in the add-in under Account & subscription, or on your dashboard at ebitdai.co). Friends who sign up with it get their first month free, and you get a free month of your plan when their subscription starts.

EBITDAI Max tier, on-demand extra usage, and Kimi runs DeepSeek too

• •New top plan: EBITDAI Max at $19.99/month ($192/year) -- everything in the Kimi+ plan, plus bring your own API keys (Kimi or DeepSeek).
• •Superseded September 6, 2026: own API keys are on Kimini too, for every provider, so the only thing exclusive to EBITDAI Max is the QuickBooks, Campfire and Stripe connectors. The claim that workbook data never transits our servers on an own key holds for DeepSeek and pay-as-you-go Kimi only; Kimi Code and Gemini calls pass through our stateless relay, which stores nothing.
• •Buy extra usage on demand: $5, $10, or $25 packs from the pane or your account, on every plan. Extra usage never expires and is spent after your monthly included usage.
• •The Kimi plan (now Kimi+) runs DeepSeek models on the same plan -- one subscription, both providers.
• •Existing subscribers keep the plan and price they signed up at.

Two plans: DeepSeek $10/month, Kimi $15/month

• •EBITDAI now comes in two plans, each with as much monthly modeling usage as a $20 Claude plan: the DeepSeek plan at $10/month, and the Kimi+ plan at $15/month running Kimi k3 -- the model that matched Claude for Excel in our benchmark.
• •Annual billing is 20% off on both: $96/year (DeepSeek) and $144/year (Kimi).
• •Both plans include EBITDAI Max: bring your own Kimi or DeepSeek key -- often cheaper per token.
• •Superseded: own keys later became exclusive to EBITDAI Max (August 29, 2026), then returned to Kimini for every provider (September 6, 2026). See the current plans on the pricing page.
• •Existing subscribers keep the plan and price they signed up at.

Kimi pay-as-you-go keys

• •The add-in now accepts metered Kimi API keys from the Moonshot Open Platform as a new provider, Kimi (pay-as-you-go) -- same k3 model as Kimi Code, billed per token ($3/M input, $0.30 cached, $15/M output), and no waitlist.
• •Calls go directly from your device to api.moonshot.ai: your key and workbook data never touch EBITDAI servers.
• •The Connect screen now explains Kimi's two kinds of keys and steers each to the right provider, since subscription and platform keys are not interchangeable.
• •Setup steps: ebitdai.co/tutorial#kimi-payg. Full announcement: ebitdai.co/blog/kimi-pay-as-you-go-api-key.

New chat and chat history

• •Start a fresh conversation anytime with the new-chat button in the add-in header -- the old one is saved automatically.
• •The clock button lists your recent chats; reopen any of them to pick up where you left off, or delete the ones you're done with.
• •Conversations now survive pane reloads: Excel Online can suspend the taskpane at will, and your chat is one click away instead of gone.
• •Chats are stored only on your device and are cleared when you sign out.

File attachments in chat

• •Attach files to the add-in chat with the paperclip, drag-and-drop, or paste: spreadsheets (.xlsx/.xls/.csv), Word (.docx), PDF, PowerPoint (.pptx), text and Markdown, and screenshots.
• •Documents are read on your device -- only the extracted text goes to the AI, the file itself never leaves your machine.
• •Screenshots work on the included plan and on Kimi Code: paste or drop an image and the agent reads what's in it, down to the numbers on screen.
• •Superseded September 8, 2026: attachments are now included on paid plans only.

Pricing update

• •EBITDAI Pro is now $19.99/month, or $167.88/year (30% off, which works out to $13.99/month).
• •Existing subscribers keep the price they signed up at.
• •Superseded August 28, 2026: the Pro plan and its $167.88 annual price were retired. The current top plan is EBITDAI Max at $19.99/month or $192/year.
• •Superseded September 7, 2026: EBITDAI Max is now $25/month or $240/year.

Relaunch: bring-your-own-key architecture

• •Own-key architecture: connect your own AI via a Kimi Code subscription key (flat $19/mo, Anthropic-format API) or a DeepSeek API key (pay-per-token, zero markup). Your key stays on your device.
• •New agentic add-in: the agent reads, writes, formats, and audits your actual workbook. Excel web and desktop, Windows and Mac.
• •New pricing: $10/month or $60/year, down from $20/month. No usage fees from EBITDAI, ever.
• •Free tier: 15 AI actions to start, no credit card required.
• •Subscribe without leaving Excel: checkout now runs inside the add-in.
• •New /addin page explaining the cost math versus bundled AI subscriptions.
• •Security hardening pass across the backend.

Older versions of EBITDAI used a credits system. It has been retired. No more buying credits, no more watching a balance drain.