Sign inCreate account

WHAT'S NEW

Mimi changelog

Release notes for Mimi: new features, improvements and fixes by version.

The last 10 releases, newest first.

0.97.9

October 23, 2026

Improved

  • Website testsReports keep recorded answers, written interpretations and incomplete visits distinct. PDF exports include the question and its answers, and prices retain their currency with unchecked budget comparisons clearly marked.
  • Website testsStarting a test and searching audiences stay responsive while audience advice or AI question writing loads. Replay cards keep names, simulation labels and progress readable on smaller screens.
  • Website testsReplays retain unsuccessful navigation attempts. Searches and page details keep the content the browser actually showed, including sites that update without opening a new address.

Fixed

  • Website testsRetrying after an uncertain connection recovers the original test and credit hold. Saved setups and repeated tests keep task settings, while signing in to a demo account always requires fresh consent.
  • CreditsFollow-up visits release their credit holds when canceled, interrupted or expired. Waiting tests and follow-ups share browser capacity more fairly.
  • Website testsSetup distinguishes the estimated cost from the maximum credits reserved. Sharing controls stay reachable in short or narrow windows, and corrected landing addresses clear their old error.
  • Website testsProduct maps count each person's recorded visits and keep separate search destinations distinct. A visit that starts on a product is labeled as starting there.
  • Website testsNamed find results check the requested item and keep other recorded choices out of completed finds. Follow-up replays and clips use the follow-up's own task labels.

0.97.8

October 23, 2026

Fixed

  • ResultsScore averages in study reports, summaries and analysis now require a complete usable answer split. Partial score bands no longer become a misleading number out of 100.

0.97.7

October 23, 2026

Fixed

  • APIAPI and MCP results leave a score average unavailable when its recorded answer distribution is incomplete or unusable. Partial answers no longer become a misleading score out of 100.

0.97.6

October 23, 2026

Fixed

  • APIAPI and MCP results leave a ranking unavailable when either its best or worst answers are missing or unusable. Missing answers no longer produce an invented order.

0.97.5

October 23, 2026

Improved

  • Feed EmulatorSetup explains the simulated audience and credit hold before a test starts. Creative previews stay with the draft as you move between steps, and keyboard controls work across its choices.
  • Feed EmulatorResults distinguish usable ad decisions, missing answers and people who left before the ad. Without calibration, they show simulated counts without calling a winner or predicting campaign performance.

Fixed

  • Feed EmulatorRetrying a start after an uncertain connection recovers the original test and credit hold. Replacing an upload keeps the newest creative, and stopping a test waits for confirmation or offers a retry.
  • Feed EmulatorFeed preparation handles sources finishing while it reads their posts, and paired tests keep the same organic posts while matching each person's interests where possible.

0.97.4

October 23, 2026

Fixed

  • APIReading an earlier study run through the API or MCP now gives that run's headline alongside its answers. It no longer borrows the latest run's headline.

0.97.0

October 23, 2026

Added

  • ResultsA folder can be moved into another folder, or back to the top, from its menu.

Improved

  • ResultsA move to a folder shows on the board at once, with Undo to put everything back where it was.
  • ResultsEach folder has its own archive: Archived counts and lists only what is archived from the folder open, and the page and its trail name the folder.
  • ConsoleThe sidebar marks the folder open and lists the folders inside it. Deleting a folder says how many items go back on the board.

0.96.0

October 23, 2026

Added

  • ResultsFolders for your studies and website tests, to keep them by brand, client or campaign. A folder can hold folders one level down, such as Nike ▸ Q4 launch. Select studies on the Studies page and choose Move to folder, or add one to a folder from its own page. Your folders are your own.
  • Smart RunA study started from inside a folder with New study here, or run again from a study that is in one, is saved in that folder.

Improved

  • ConsoleYour folders are listed in the sidebar, and the way back from a study opens its folder.

0.95.3

October 23, 2026

Added

  • APIrun_study, ask_study and ask_followup take an idempotencyKey: a retry after a timeout gets the first run or answer back and is not charged again.
  • APIlist_studies pages through all your studies with cursor and nextCursor, not only the newest 100.

Improved

  • APIA key made on the Connect page for an agent can read and run but not make public links, and spends at most 5,000 credits until you raise its cap in Settings.
  • APIThe MCP server lists each tool once by its current name, answers more compactly and says how long to wait before a retry. The Codex setup now allows the five minutes a study with media can take.

Fixed

  • Smart RunStudies opens again for everyone. The estimate of how long a run will take had grown too large to read, and its failure took the whole page down with it; it now reads a short record kept as each run finishes, and the page carries on even if the estimate is unavailable.
  • CreditsStopping a run charges for the answers it already made and returns the rest of its hold. Nothing is charged when no answer came back.
  • APIRunning a study again through the API uses the settings the Run button would, so a study whose cohorts cannot pass influence between them runs again instead of being refused. Follow-ups read the last finished run while a re-run is going or after one stopped.
  • APIshare_study links only to studies with results, run_study asks you to restore an archived study first, and a failed run's reason no longer shows internal error text.
  • AccountChanges to an organization's members are applied as Clerk has them now, so a delayed update can no longer add back someone who was removed or reopen a deleted organization.

0.95.2

October 22, 2026

Fixed

  • AccountAccepting an invitation to an organization brings you back to Mimi as a member straight away, instead of stopping on a sign-in page. A new account invited by an organization's administrator no longer waits for approval.
  • AccountSomeone invited to your organization who already has a Mimi account is told in the bell and can join from the new Invitations page. Its administrators are told when they join.