What's new on ToRun
No marketing copy, just the diff — newest first.
August 2026 40 pages of history
- FixedThe channel editor stopped asking you to type engine names Six fields in the mobile channel editor were plain text boxes — output language, narration voice, and the picture, video and music engine names — so filling one in correctly meant knowing an internal identifier, and a ty…
- FixedA Russian to Turkish translation came back in English The live translator was still tied to one fixed engine, so it never reached the part of the system that picks a model on quality and price — and that engine answered in English whatever language was chosen underneath. Th…
- FixedA run is quoted on the models it will actually use The estimate shown before a run counted only the steps whose price was already known, so a run quoted at seven percent of a session budget could stop dead at the fourth step because one picture cost three times that — an…
- FixedA pack the store sells is a pack we can credit A top-up pack was on sale on Android with no matching entry on our side, so the purchase went through, nothing arrived, and the store refunded it days later. Worse, accounts that hit this were being filed as suspected ab…
- FixedThe slideshow moves, and its seams fade instead of cutting A single still stretched over three minutes moved too slowly to read as motion at all. Stills are now split so each gets a visible move, and two or more of them are joined by a crossfade rather than a hard cut. A fade ta…
- AddedReport anything the assistant makes, from anywhere it makes it A report control now sits on every surface that generates something — chat replies, pictures, video, music, voice, speech to text, text to speech, translation, canvas, workflow results, the assistant bubble and shared pa…
- FixedWorkflow forms in your language, with dropdowns that are not blank Field labels on run forms were English for everyone, and putting a Turkish word in capitals turned i into I instead of İ. Both are fixed, on the web and on the phone. Separately, 51 templates preselected a value the drop…
- Fixed“Charge my wallet” wrote nothing, so the turn asked again On the chat page, answering a spend question by choosing your wallet was rejected by the server without a word, so the turn replayed, asked the identical question, and failed — no paid turn could be completed there at al…
- FixedThe code step could never run — now it does, with more than one file Internally the step accepted three language names while the dropdown offered twenty-eight, so almost every choice was skipped with a message naming languages that were not on the list. It also choked on the fenced code b…
- AddedA finished run reads like a report, not a data dump The result page printed every internal variable as a row: raw names in monospace, a 14,000-character block of escaped text next to an empty list. Now the written report comes first as formatted text, retrieved sources ap…
- FixedThe multi-source research template failed on every run Its cross-check step refused any input longer than 8,000 characters — a limit written for a one-line brief, not for the search results that step exists to compare — so the template had never once produced a report. That…
- AddedA charge can now name the conversation it came from Two kinds of spend never reached the chat's own total: a music track that finishes minutes later on a background worker, and a render that failed but was still charged. Both are recorded against the conversation now, so…