What is new on the ToRun platform, newest first.
June 2026 — 53 changes
The model picker now includes Google Imagen and Gemini for images, and OpenAI Sora and Google Veo for video, alongside the models you already had. Pick exactly the one you want, or let Auto choose the best fit for your p…
Chain several AI steps into a workflow you can run on demand, then publish it to the Marketplace or browse what other creators have shared. The Workflow Editor, Runner and Marketplace are now available straight from the…
Add ToRun to your home screen and launch it like a native app, full-screen and with its own icon, so it is one tap away until the mobile app arrives.
A new logo and cleaner branding across the whole app, tuned for both the light and dark themes.
Choose how hard the AI thinks, now even in Auto mode. Image prompts can quietly draw on your saved memory so results match the people and details you care about. And near-duplicate memories are merged automatically inste…
Generation progress now appears right beside the result instead of under the form, each page opens clean, and your latest creation stays front and centre while older ones live in your Library.
Your wallet history is now paged and shows exact sub-cent amounts, and removing items from your Library no longer throws an error.
Create your advertiser account and submit campaigns for review straight from your account, with no waiting on manual setup.
Behind the scenes: fairer free-credit accounting, stronger checks on creator payouts and permissions, safer media moderation, and more reliable workflow runs.
Live Translation now lets you choose between two engines and pick the one that fits your conversation — including a new option that speaks 77 languages, Turkish and Russian among them. Each choice shows its price up fron…
Models your plan doesn't cover now appear with a small lock and a one-tap way to upgrade or top up your wallet, instead of failing only after you send. We also opened more capable models to the Free plan, so everyday mod…
Share a picture once and the assistant keeps what it saw for the rest of the conversation — so you can ask about it many turns later, even after switching models, and it never re-uploads the image or charges you twice.
Chat answers now draw diagrams as real diagrams, lay out tables with clear borders, and size headings to fit the message instead of filling the screen — so replies are easier to read at a glance.
When a model needs an upgrade or your balance runs low, you now get a friendly banner with Upgrade plan and Top up wallet buttons right there. Replies that can't be generated explain why instead of leaving an empty bubbl…
Billing records, history and activity now display in your local time rather than UTC, so timestamps match the clock on your wall.
Starting a new chat now reuses an empty one instead of stacking blanks, chats that never got a reply no longer clutter your list, and placeholder titles are replaced by a real name as soon as you say something meaningful…
Businesses can now set up an advertiser account and run their own ad campaigns end to end — create campaigns, add the ad creatives that go with them, and submit them for a quick review before they appear.
Bringing your own provider key now covers realtime voice, live translation and live transcription too: the AI usage goes through your own key and we only charge the small per-minute platform fee.
Our public pages are now driven by editable content in many languages, uploaded images are scanned for unsafe content before they are stored, account sign-up and confirmation e-mails work reliably, and a broad round of b…
Long replies now run on the server and carry on while you switch chats, open another page or close the tab. Come back any time and the answer is still streaming, or already waiting for you.
Chats started inside a Project automatically pick up its instructions, uploaded knowledge and memory, show a project badge with a quick link back, and name the project in your recent chats. You can also add PDF, Word and…
Your assistant's memory now tracks how quickly each fact changes and flags the ones that may be out of date, so it leans on what is still true. A new editor lets you review, add and remove individual memories yourself.
We have added the latest image, video, transcription and realtime voice models, and opened standalone Translate and Transcribe pages so you can reach those tools straight from the menu.
Speak and have your words translated almost instantly: choose two languages, swap them with a single tap, and the translator captures your voice and streams the translation back as you talk.
The video page now keeps its tiles in step when you delete from your Library, lets you set the clip length per model, remove individual clips, and clearly marks preview results so you always know what you are looking at.
The in-app assistant now shows nicely formatted replies, keeps the latest message in view as it types, reliably saves the finished answer, and gives each conversation an automatic title.
Replies that contain links, mathematical formulas or currency amounts no longer show up blank or half-rendered, even after a page refresh, and deep-research answers always include their final summary.
A broad round of behind-the-scenes work: more accurate usage metering and spend totals, fairer automatic refunds, sturdier real-time messaging, stronger privacy and security safeguards, and clearer service-status message…
Public blog posts and forum discussions — including their reactions and replies — are now visible to signed-out visitors.
Usage is now charged with a single, fair markup, your monthly-spend and remaining-budget gauges read accurately, and the model list has been cleaned up — dead models retired and new flagship models wired in.
Workflow listings and bundles can display a proper cover image, making the marketplace easier and more pleasant to browse.
Ask the assistant and it will fill in forms, click buttons and move around the app for you — backed by a broader toolset that now reaches contact, the FAQ, the forum, memory and one-step workflow creation.
Answers now stream in word by word instead of appearing all at once, and the assistant bubble can be dragged wherever it suits you, with a live hint of what it is working on.
Creating a project and starting a new chat inside it now works smoothly, so you can keep related chats, files and context together.
Videos play reliably, private media loads through a secure proxy, creator and persona avatars now show for everyone, and voice notes are transcribed without errors.
When part of the service is having trouble, the notice is now easier to understand and no longer exposes internal vendor names.
Link external apps with a single sign-in or an API key, so the assistant can work directly with the tools you already use.
Image tools now pick a model that matches what you are doing, show an instant inline preview of the result, and editing works again after a provider model change.
When an image or attachment has been removed, your chat history, media gallery and image editor now show a friendly placeholder instead of stopping with an error.
The translator now offers a searchable picker with roughly 100 languages, each shown by its native and English name.
We have added a batch of new blog articles and forum threads — including answered questions — so there is more to read and explore from day one.
Chat can now search your uploaded files, run Python in a sandbox, edit images, transcribe and speak audio, and trigger your workflows — all without leaving the conversation.
Let chat search the live web, run multi-source deep research, and fetch and read full pages — every answer comes back with the sources it used.
Your assistant now keeps lightweight memory across chats, can list what it remembers, and honours a forget request in any language.
Generate and edit images, produce short films, and compose songs with lyrics — powered by new image, video and music engines, with a shared media picker and your library one click away.
Talk to ToRun out loud — a new realtime voice capability brings spoken conversations and live transcription to chat.
Workflow runs now prefer premium-quality models, pause for your approval where it matters, and can be cancelled, regenerated and resumed — with every output saved straight to your library.
Group related chats, files and context into Projects so your assistant always has the right background for the task at hand.
Draft and iterate on long-form content and code in a dedicated canvas panel right next to your chat.
Rolling summaries, smarter context windows and faster history scrolling keep long chats coherent and snappy, with clear controls when a conversation outgrows the model window.
A humane free allowance that refills over a rolling window, clearer spend gauges, automatic refunds when an image or video generation fails, and reasoning-effort controls that only unlock what your balance can cover.
Dozens of 2026 models added and stale ones retired, plus a daily auto-sync that keeps provider models and pricing up to date — managed from a new admin catalogue.
Publish a conversation to a public link and grow the community around what you create.
May 2026 — 499 changes
Standalone .NET console harness (test/ToRun.SmokeTests) that authenticates against the running Host (OpenIddict password grant) and exercises every core capability end-to-end over the real HTTP API: auth/me, memory CRUD,…
Platform media lives in a PRIVATE Bunny Storage Zone, so a plain
Completes the chat-driven memory management feature on top of
memory_forget returned no_match for memories that exist.
gemini-2.5-flash started returning HTTP 400 on thinkingBudget=128 (the 52-N low
The two host-admin pages (/admin/model-catalog, /admin/model-onboarding)
Discovery matched only by exact ProviderModelId/ModelKey, so dated provider
Reasoning-effort gate is now funding-based, not tier-based.
Best-of-both merge into the landed 52-Q-MODEL-SYNC-JOB sync: when the OpenRouter
Reasoning-effort selector was hard-gated on subscription tier plus an admin/host exemption from an earlier, rejected approach.
Phase 1 deliverable: 26-provider list/pricing-source inventory, the
The onboarding wizard (phases A-D) only CREATES catalogue rows; there was no
Backend (torun.agent): SharedChat gains IsListed/ListedSlug/ListedAt + denormalized card fields; PublicCommunityAppService + PublicCommunityController (anonymous /community feed + by-slug viewer); CommunityListingAppServ…
Audio attachments in chat were broken; the actual root causes differed from
PART 1 — the chat-header "Context overflow handling" dropdown was stuck on
The daily-cap removal (84178afab) deleted DailyKey/MonthlyKey from BudgetBucketCacheItem
Part 1 — un-red the MongoDB integration-test floor.
Reasoning-effort selector strings (Effort:Tooltip, Effort:Locked:Tooltip,
Canonical, provider-agnostic reasoning-effort control on the chat composer,
Final dead-code pass for the session-window budget rewrite.
The Monthly/Daily Spend gauge reads telemetry UserDailySummary (ScopeKey=userId,
Audit follow-up to the session-window budget rewrite (2a33fd31f).
Home (/) + me-home budget-bucket gauge rendered "$0.00 / 0% / $0.00" whenever
Free-tier budget bucket was a "$2 / 30d / 24h" drip: refill = $2/720/h and
Home (/) and me-home (/me) BUDGET BUCKET gauge showed "$0.00 (2% / $0.01)":
Multi-turn vision bug.
First regression-lock in the smoke layer.
Root-cause fix for the recurring title-gen / public-chat / vision 404s: capability
User confirmed image description now works.
Owner-scope 403'd even for the legitimate chat owner: manager-created
A3 REAL root cause (found by live browser test): chat-attachment image
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Resolving IQueuedRequestDispatcher StackOverflowed at startup (uncatchable — a
Both the Host runtime sink (appsettings.json) and the bootstrap loggers
Recurring root cause: ChangelogBackfillDataSeedContributor probed idempotency on
Ran extract-sprint-changelog.py + propagate-changelog-locales.py from main:
QueuedRequestRunner.DoWorkAsync resolved IQueuedRequestDispatcher at the TOP of
The chat overrun check compared raw history tokens against the TIER budget (ctxBudget.MaxInputTokens) only, so when the tier budget exceeded the routed model's context window the trim was skipped: e.g.
Proactive "concern toast" (Honest-UX DNA pattern #3).
Translate the 16 Chat:Context:* (Agent resource) overrun-prompt/footer/toast keys and the 13 MySettings:ContextOverrun:* (ToRun resource) settings keys from en.json into all 28 sibling locales (en-GB verbatim).
Propagate the EN-only humane-error + status-banner keys (ErrorHumanizerService
Phase B surface for the chat-canvas (ChatGPT Canvas / Claude Artifacts
Honest-UX DNA pattern #3.
Completes the wallet-funded context-overrun feature.
Propagates LEARNING-LOOP Phase 2 keys (Admin:UserExperience:ComplaintLoop:{Tab,Title,Subtitle,FixedInWindow} + Admin:UserExperience:TimeToFix:Empty) plus 5 earlier EN-only Permission:UserExperience.* strings from the sam…
Four parallel 52-Q chips landed code referencing symbols their sibling chips
Closes the 29-locale parity gap for the CHAT-CANVAS / ChatSummaries slices,
Phase 2 Scope B of the honest-UX-DNA pattern #5 loop.
Honest UX DNA pattern #2 (Reversibility) — make the auto-refund VISIBLE.
Honest UX DNA pattern #2 (Reversibility) — wire the projection layer
Honest UX DNA pattern #2 (Reversibility) — refund the user when something
Honest UX DNA pattern #2 (Reversibility) — user-facing read API for the
EN-only this commit — the paired slice52-Q-AUTO-REFUND-LOCALE chip fans
Honest UX DNA pattern #2 (Reversibility) — append-only audit row for
Paired locale fan-out for the /me/projects feature (components landed in
EnBillingSource.RealtimeVoice + voice-realtime capability seed + join
Backend exception → frontend humane message + retry hint + status banner.
Consolidator landed both 52-Q-CHAT-CONTEXT-BUILDER's ChatSummaries
Mirror of the realtime voice-mode UI keys added to en.json in
Native translations of the 40-key User Sentiment / Privacy Policy v3 Section 11 block (legal text per GDPR Art.
Suite-managed file edits (Permissions+MongoDB+ChatPipeline)
land additional parallel-chip WIP
Consolidates the working-tree state stashed during the
Adds IChatSummarizationService + concrete impl that compresses the dropped
Third of the four P0 chat tools — ChatGPT File Search parity.
Why: backend tier-budget context builder (slice52-Q-CHAT-CONTEXT-BUILDER)
Adds ImageEditTool (mirror of ImageGenerateTool) so the LLM can edit user-
Fans out the 46 Agent WebSearch + Menu:WebSearch + Permission:WebSearch.*
Why: P1 voice tools listed in feedback_chat_context_management.md — text
Closes the user-complaint → shipped-fix loop automatically so we never make
Backend:
The /me/web-search slice landed almost entirely via Serkan's recent
Post-streaming runner that drives Pyodide in a Web Worker and feeds
MyMediaUploadAppService.PurposeChatAttachment validation extended to
Two-endpoint AppService that bridges the server-side code_interpreter
Ship the ADR-043 Pyodide v1 backend slice — the LLM-facing IAgentTool
Why: long chats (100+ turns) bloat browser memory + slow ngx-markdown re-renders
Fourth of 4 P0 chat tools per feedback_chat_context_management.md roadmap and
Phase 1 design spike for code_interpreter chat tool (second P0 tool after
Public-chat launcher now boots a Cloudflare Turnstile widget at view
Locale key for the new Cloudflare Turnstile rejection banner shipped
The [AllowAnonymous] PublicChatAppService now verifies a Cloudflare
Marketing landing's anon chat now streams the assistant reply
Surface the three new WebSearchOptions knobs in appsettings.json so the
Three tools wired to the canonical authoring-tier IWebSearchProviderFactory.
Catalog seed for the dual-tier chat web-search tooling.
Adds the second half of the provider contract needed by the dual-tier chat
User report 2026-05-27 — public chat marketing surface 60s+ bekleyip
User: "ana sayfada neden markdown renderer yok, direk markdown olarak
Bridges the gap between DeepResearchSession CRUD (sat at default state forever)
Propagate Public:Chat:Error:NoModel / CallFailed / EmptyPrompt
User report 2026-05-27: pricing sayfasındaki Partner card'ın
User: "burayı geçen gün yaptık sağlamdı, yine bozuldu.
Slice 452-fu already shipped first-class Brave + Tavily IWebSearchProvider
Replaces the stub web_search tool with a config-selectable provider:
User: "deepseek response geldi ekran 2'de, sayfayı refresh edince
Native-quality translations for "Action completed.
User şikayet: "gpt 4o mini response boş geldi.
User: "her sayfa refresh ettiğimde chatlerin yerleri ve sıraları
Önceki commit (e2a74db8) sadece relativeTime/fullTimestamp helper'ını
Sprint 50-S backend fix made the wire format emit `Z`-suffixed ISO
User direktif: "arkada iş kalmasın".
Önceki "fix" (a56449db) Opus 4.7 / Sonnet 4.6 için preemptive olarak
Pre-fix rendered the warning triangle as an inline <i> followed by
Önceki rev (a56449db) sadece ilk-mount fallback'i çözüyordu —
Third regression of the same bug class in three sessions.
User reported four issues in a single round of chat smoke-testing:
Atladığım şey buydu.
Three iterations of the same bug (Monthly Spend $0 → $0.33 → $0.15) had
Mic-capture via MediaRecorder (chat composer Whisper STT, Sprint 52-O)
OpenAI gpt-4.1 family (Deprecated → Retired, isActive:false, version 3→4):
Close 10 pricing rows for the 5 retired models:
Anthropic (DeprecatedAt 2025-09-29, RetiresAt 2026-05-01):
User reported a wide swath of dropdown rows error on send:
User report: "reason yapar gibi baya bekliyor ama hareket yok, reason
Audit continuation after slice52-Q-FINANCIAL-AUDIT-1.
- PROGRESS.md: Sprint 52-Q-USER-BILLING-DAILY entry with all 4 commits,
User report: "monthly spend 0.33 olamaz, burada bir hata var, muhtemelen
- New IDataSeedContributor: UserBillingDailyBackfillDataSeedContributor
- New IInboxConsumer: GroupId "torun-user-billing-daily-summary"
- Suite entity JSON: modules/torun.subscription/.suite/entities/UserBillingDailySummary.json
User: "TelemetryUserDailySummaries tablomuz var gördüğün gibi içide dolu.
User report (continued from slice 52-O-NIGHT-TODO-IMPL/home-spend-fallback):
User report: "monthly spend her gün sıfırlanıyor — redis recycle ettiğimden,
activity routes done + remaining TODO inventory
User feedback: "kod içerisine daha önce yazdığımız TODO lara hiç
TODO cleanup audit ledger
User directive: "TODO ve benzeri yorumları ara, sonraya bırakılmış iş
morning report for Serkan — overnight locale + changelog ledger
These four ad-hoc scripts drove the overnight locale recovery:
extract-sprint-changelog.py wrote 364 sprint rows to
Followed up the placeholder fill (commit a3834ea4) with native-language
Audit found 1311 missing keys spread across 27 sibling locales (en-GB
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Adds the 11 EN-master entries the new SourceContext chip strip
Surfaces the new SourceContext metadata as a user-facing filter so
Closes the chat-attachment data-loss bug: previously the composer
ChatTurn interface gains audioTranscripts[] with an expanded flag.
Add CapabilityKeys.AudioInput ("audio-input") so the fallback condition is
The chat composer's prior path inlined image attachments as base64 data
Pre-fix, the composer cost badge only sized the draft text via
`:streamGenerateContent` without `alt=sse` returns a JSON-array transport
GetActivePricingAsync now falls back to model-level pricing rows when the
Adds 9 new Chat:Continue:* keys (Button, Tooltip, Loading, Error, MaxReached,
Angular proxy: +wasTruncated/continuedFromMessageId/hasContinuation on both DTOs;
OpenAI Chat Completions silently consumes reasoning tokens for the
PersistAssistantMessageAsync now sets FinishReason/WasTruncated/ContinuedFromMessageId
Extend ChatMessage.Extended.cs with 4 new virtual properties (FinishReason,
User reported the "Daily free limit reached — Charge wallet" yellow
Propagate the new MediaRecorder + Whisper STT composer keys from en.json to all 28
Why: legacy Web Speech path produced transcript-only with no audio artifact
User reported two chat UX papercuts after the chat-footer fix landed:
User-reported chat composer trio:
Propagated Changelog:Sprint:<slug>:Title and :Body keys for the 38 new sprints
Minor bump for the Phase 1 feature wave summarized in the changelog backfill
Ran extract-sprint-changelog.py against HEAD.
- New IChatPipelineAppService.GenerateTextAsync endpoint (creative-writer
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Comment introduced in 57c15021 used backticks around '/marketplace' which
**Bug 1 — Card titles render raw `WorkflowTemplate:foo:Title` keys:**
Per-step output variable snapshot so the resume-aware wave loop can
In-memory wait loop in WorkflowExecutorAppService.RunWaveLoopOnRowAsync (line 1797) is
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Inject HumanApproval (kind=12) node into all three birthday catalog
WorkflowTemplate:BirthdayCinema:Title + :Description propagated to all 26
User wanted the third leg of the birthday trio: "isteyen kullanıcı video
Birthday workflows expose a textarea field 'provided_lyrics'; when filled
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
THE photo-binding bug.
User observation: "library bakınca sadece image görüyroum ne ses var ne
User report: merge step "Succeeded" but composite_url_duration_seconds
User report: "succeeded yazıyor da bunlar success degil ki hata veriyor
User directive 2026-05-25 — region prefix produced 401 Unauthorized on
Root cause (from Note column added in 0a9e941f):
Root cause: ImageCapabilityRunner with n>1 packs all generated URLs as a
Adds an "Or pick from your library" link/button beneath the workflow-wizard
Replaces the /me/audio direct file input (audio.component.html line 17 —
Replaces the /me/image Edit-tab mask widget (bespoke library-only modal +
Replaces the inline "Drop an image or click to upload" widget for character
Universal two-tab picker that replaces PC-only uploads across /me/* surfaces.
Pre-fix VideoPipelineAppService.GetCapabilitiesAsync (commit 5e39d434)
Pre-fix /me/video catalog query returned only sora-2 + sora-2-pro because no
Native translation per locale (prefix only; the numeric seconds cap
Pre-fix the duration slider on /me/music was hardcoded max=120s.
Pre-fix VideoPipelineAppService.GetCapabilitiesAsync returned a single
User generated music with `requestedDuration=120s` against Riffusion
User got
User got the UFE
User got HTTP 401 from Replicate after triggering musicgen and the
User reported the /me/music model dropdown stayed empty even with
Console showed:
Carry-forward from 05999c1c.
User reported video + music menus missing from My Account sidebar even
Mirrors fc625886 EN seeding into ar/cs/de/de-DE/en-GB/es/fi/fr/hi/hr/
Angular standalone component mirrors /me/voice + /me/image shape:
User-facing music generation pipeline mirroring ImagePipeline pattern:
The AuthoringPermissions.VideoPipeline.MyGenerate comment says the
Second-pass merge failure after b9dm94er4 + 21983ef0 still surfacing
Adds the Sprint 52-Q ModelOnboarding admin UI strings (Menu:ModelOnboarding +
Fans out the 32 EN keys added in commit be0190b2 (admin /cost-markup
39 MR keys (Menu:Admin:MarginReconciliation, Permission:MarginReconciliation, MarginReconciliation:*) added to every sibling locale under src/ToRun.Domain.Shared/Localization/ToRun/.
Sprint 52-Q-MODEL-ONBOARDING Phase D — round-trip bulk-import / bulk-export
Sprint 52-Q-MODEL-ONBOARDING Phase C — host-admin Angular UI at
Guardrail sweep finding from Phase D of the 52-O-MARGIN-AUDIT sprint.
Sprint 52-Q-MODEL-ONBOARDING Phase B.2 — adds the GenericRestClient that
Phase C UI of the 52-O-MARGIN-AUDIT sprint.
Sprint 52-O-REAL-COST-AUDIT phase C — each pipeline AppService that wraps
Sprint 52-Q-MODEL-ONBOARDING Phase B.1 — scaffolds the AiProviderClientStub
Phase C of the 52-O-MARGIN-AUDIT sprint.
Sprint 52-Q-MODEL-ONBOARDING Phase A — adds AdminModelOnboardingAppService
Sprint 52-O-REAL-COST-AUDIT phase B — every IAi*Client now stamps a
Sprint 52-O-REAL-COST-AUDIT phase A — parallel server-side-only ledger so
Phase B of the 52-O-MARGIN-AUDIT sprint.
Phase A of the 52-O-MARGIN-AUDIT sprint.
Error from DbMigrator on every restart:
Carry-forward from b4ab9daf music_v1 seed bump + feedback_pricing_
User feedback after 21983ef0 landed: looked at ElevenLabs analytics
- ToRun resource: 36 keys (Library:Share:* dialog + SharedMedia:* public page)
Pre-fix: workflow ArtifactSave nodes uploaded blobs to Bunny and stuffed
Sprint 51-M-fu2 completion pass.
Birthday-song-video template's `photos` multiphoto field reached Bunny CDN
Drops native v2.0 About/Terms/Privacy/Contact for 10 brand-new locales
EnhancePromptAsync wrote a BillingRecord and settled the budget but never
Root cause (two layers, one bug):
LoadHistoryAsync previously walked the leaf → ancestor chain by issuing one
Per CLAUDE.md Rule 10.
Output of `python platform-plan/scripts/_validate-29-locales.py` after all 27 sibling-translator batches landed + the BOM-strip pass:
Batches from 6 parallel Sonnet translator agents (all checked into the repo from earlier turns + now-flushed final group):
Serkan caught this — the Generate tab on /me/image has a card-shaped progress bar that walks through Preparing → Requesting → Generating → Fetching with a striped animated `progress-bar bg-info`, phase-specific icon + la…
The bug Serkan surfaced from /me/workflow-runs (Detail click on a Failed run): every workflow step that delegates to an ApplicationService with class-level [Authorize] (ImagePipelineAppService, all the pipelines) termina…
Completes the locale-fanout begun in 0da3c077 (Sprint 52-O-12-locale).
Read-only audit pass against ChatPipelineAppService + InAppAssistant + provider clients + ProviderFallbackChain.
7 of 19 stub locales populated; 12 remaining (cs, sk, hu, ro-RO, hr, sl, ko, zh-Hans, zh-Hant, hi, ar, vi).
Sprint 52-O-18 added ModelLifecycleStatus.LegacyAvailable (used to mark Claude 4.5/4.1/Opus 4.6/Sonnet 4.5 etc.
Sprint 52-O-19 (provider runtime hardening) complete — 7 commits land:
Concrete unblock: AiModel rows whose ProviderKey was set to "musicgen" / "ltx-video" / "flux-1.1-pro" / "hunyuan-video" / "sd3.5-large" / "xtts-v2" / "bark" all route credential lookups through "replicate" when no per-mo…
New IProviderCircuitBreaker (Application.Contracts/AiClient/Resilience/) + default ProviderCircuitBreaker impl backed by IDistributedCache<CircuitState,string> keyed by torun:byok:circuit:{providerKey}:{capabilityKey} wi…
# Conflicts:
ProviderKey="elevenlabs".
ProviderKey=\"replicate\".
PayloadGrammarJson column declared on AiModelCapability.Extended.cs — backports the Sprint 52-O-17 chip's column that the worktree branched off main without (worktree base is 16 commits behind local main, so 52-O-17 wasn…
Adds ~60 new pricing rows in the 520180XX-...
ResolvedAiProvider DTO carries the read-only snapshot that generic adapters consume at request time — protocol drives which adapter class handles the request, BaseUrl is forwarded into the outbound HttpRequestMessage so…
SeedSlice52O18Async() runs at the end of SeedAsync regardless of count gate (idempotent EnsureCapabilityRowAsync calls, deterministic XOR-based GUIDs so reruns are no-ops).
Existing 108 rows: status / deprecatedAt / retiresAt patched per the deprecation calendar in 00-MASTER-SUMMARY.md §"Major Deprecation Calendar" — gpt-4o / 4.1 / 4.1-mini / 4.1-nano / o1 / o3-mini / gpt-image-1 land 2026-…
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
New shared enum EnWireProtocol (11 named + Custom catch-all) covers the realistic provider population: OpenAiCompat (OpenAI/xAI/DeepSeek/Mistral/Groq/Cerebras/Together/Perplexity/Azure-OpenAI/Fireworks/DeepInfra/OpenRout…
EnqueueRawSpecAsync inserts a Pending WorkflowExecution row, stamps QueuedAt + Activity Center row, emits torun.authoring.workflowexecution.queued.v1 via transactional outbox (Rule 8), and returns the DTO so the wizard n…
Claude 4.5 / Opus 4.1 / Sonnet 4.5 / earlier Gemini 2.0 / xAI Grok 3 are all examples of models that stay on the API surface but the provider is steering new traffic to a newer release.
EnPricingUnit grows 12 → 15 to model 2026 billing meters the legacy enum couldn't represent.
The 2026 model generation introduced capabilities that the old 18-key vocabulary couldn't address: Claude's adaptive vs.
23 → 25 providers (stability re-introduced after slice 4xx removal, two OSS-serving newcomers fireworks + deepinfra added).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Pre-52-O-15-fu the OpenAiImageClient hand-coded every model-specific quirk inline.
Backend (fu1): new POST /api/subscription/budget-preview/preview wraps IBudgetGuard.CheckAsync read-only.
OpenAI image API rejects unknown sizes outright with HTTP 400 + "Invalid size 'WxH'.
WorkflowSelectOptionsDataSeedContributor now does an idempotent per-(option, locale)
User feedback on Sprint 52-O-13: "tabi bunun için kullanıcıyı bi popup ile uyarmalıyoz." Letting the wallet bypass the daily cap is the right behaviour for paid credit (52-O-13 fixed the silent reject), but the user stil…
The Free-tier $0.10/day cap exists to spread $2/month allowance across days so a single user can't front-load the host's AI-cost reimbursement.
Sprint 52-O-12d completes the catalog story started by 52-O-12a (entity)
Hygiene-only follow-up to 52-O-12a/b.
User added a per-locale enrichment block to
Sprint 52-O-12a follow-up — DI bootstrap fail
Phase B (Angular wizard) for the 52-O-12a data-driven dropdowns.
Phase 1 backend stack for data-driven workflow wizard dropdowns.
Backend bug — Cache Manager shows 3 categories of ~10+, every Browse empty
UX bug — cryptic exception copy on /media/video-stream
Backend bug — empty Recent transactions list with non-zero balance
UI bug — multi-theme breakage
Backend bug — entire workflow editor surface ships raw keys
Backend bug — every workflow template run failed
Exception screenshot from Serkan (2026-05-23):
Serkan's motto (2026-05-23): "biz 25 farklı AI provider destekliyoruz...
THE ACTUAL ROOT CAUSE of the seven "title-gen yine bozdun" reports —
52-I excluded DeepSeek V4 from cheap-model picks, but on a deployment
50-LLL-2 added an ExcludeReasoningModels filter so the cost-saving
User reported VS Error List showing CS1501 "No overload for method
F15 (root cause: AgentAuditEntry table unused for the bubble's
Four backend regressions in InAppAssistantTurnAppService surfaced
Audit #14 — confirm Sprint 51-A's reasoning opt-in path is per-request,
Two more console warnings flagged in /me/chat after 52-E:
Bundles four small chat-pipeline polish fixes from the audit (#10, #11,
F10 (root cause: bare WriteSseEventAsync in the controller's outer
Three UX / a11y polish items from the sprint 52-D code-review pass.
Two DOM-side hygiene fixes from the sprint 52-D code-review pass:
Three concurrency / state-coherence regressions in the bubble component
Three SSE-stream regressions in the InAppAssistant bubble shipped with
Three chat-pipeline surfaces were emitting hardcoded strings in violation
Serkan flagged ~148 identical "The localization source separator (::)
C5: ResolveFallbackProviderKey returned the literal string "google"
Every state write in chat-pipeline.component.ts goes through a Signal
Three lifecycle / concurrency defects in the chat-pipeline component, all
Serkan flagged a visible dead band between the composer card and the
ChangelogBackfillDataSeedContributor reads from the embedded resource
> "bu contact formunda bütün maillerimizi açık açık yazdık spammerlara mail
Follow-up to Sprint 51-V (commit 373d84d9) which rewrote en.json only.
Smoke at 2026-05-22 showed `/personas` listing "Echo the Storyteller"
Three compounding bugs from 51-S → 51-U all visible in the same screenshot:
51-S / 51-T drifted from the catalog-first vocabulary and shipped a
Cleanup of regressions introduced by 51-S.
Sprint 51-S (28e76a7d) shipped lib-pricing-grid with a contact-sales price cell that
ONE pricing surface for all 6 plans (Free / Starter / Pro / Business /
User mandate (2026-05-22): "Bu description'ları biraz zenginleştir.
User mandate (2026-05-22): public surfaces leaked the visitor to a new
User report (2026-05-22): "Login durumdayken pricing geliyor, public
Two user-flagged issues on the same screenshot (2026-05-22, anon
User report (2026-05-22): "Login durumdayken pricing geliyor ama public
User report after 51-N landed: "DbMigrator çalıştırdım, proje yeniden
User report (2026-05-22): "ABP bunu hala 20 dil görüyor ve dilleri açtığımızda 20 dil çıkıyor.
User correction (2026-05-22): "bizim platformumuz 29 dilde." Legal seeds
User directive (2026-05-22): About too thin, Terms must hold up against
Three connected anasayfa bugs landed in the same commit because they all
User mandate from 2026-05-22: "release çıkmadan önce change loglarının seed
Three connected bugs from the anasayfa list (platform-plan/BUG_LIST_ANASAYFA_2026-05-22.md):
BUG-3+4+5 from anasayfa list closed.
Follow-up to 51-C — my own mistake.
BUG-1 closeout from anasayfa bug list (platform-plan/BUG_LIST_ANASAYFA_2026-05-22.md).
BUG-2 closeout (3-part user report + 1 rule refinement):
BUG-1 part 2.
- Workflow editor, wizard, templates, validation, preview, run history strings
- Workflow editor, wizard, templates, validation, preview, run history strings
- Workflow editor, wizard, templates, validation, preview, run history strings
- Workflow editor, wizard, templates, validation, preview, run history strings
- Workflow editor, wizard, templates, validation, preview, run history strings
- Workflow editor, wizard, templates, validation, preview, run history strings
- Workflow editor, wizard, templates, validation, preview, run history strings
- Workflow editor, wizard, templates, validation, preview, run history strings
Closes ADR-042 — previously Proposed, now Accepted.
`col-md-1 text-end` clipped the localized "Clear all" label at
Sprint 50-PPPP audit — most MessageKey expressions across the codebase
TEST_PLAN §4 deferred — capability key string literals (text-chat,
- persona-profile.component.ts: drop outer `lpx-content-container py-4` wrapper
Three deferred items landed (or were dismissed) since the previous closeout
Completes the slice50-workflow-engine audit pass for the QualityGate
Continues the slice50-media Phase1Note translation pass started by
Replaces 8 [en] placeholder Admin:StorageProviders:Phase1Note:* and
Sweeps the torun.media Localization resource for the 12 non-EN locales
Replaces ~110 [en] placeholder values in the Authoring resource's
[Cursor-aware var insert (slice50-S)]
Closes the loop on slice50-I's per-node budgetUsd intent layer with
User-driven polish on the /media/storage-providers surface:
Adds OptimizerOptions sub-block under BunnyCdnOptions for Pull-Zone-level
Closes chip #118 last deferred item:
Backend (modules/torun.persona):
Closes the deferred item from chip #118 wrap doc:
Two inspector enhancements that close the gap between author intent
WorkflowGraphComponent:
Closes the deferred item from chip #118 wrap doc:
Closes the deferred item from the workflow-editor wrap (chip #118):
Two related extensions to the advanced editor surface:
Rule 8 (transactional outbox) — after the executor's RunSpecAsync returns
StorageBrowse and UploadAsset landing card descriptions had leftover
The existing executor's QualityGate node (Slice 449/449b) already implements
Single AppService surfaces three endpoints:
Workflow-node-shaped wrapper over Sprint 50-LLL's IModelRoutingService.
User-facing wizard-catalogue endpoint.
Captures what landed this sprint (library landing + simple-mode wizard
Auto-propagated from the en.json additions in 9b95c79c (admin panel
Replaces the existing add-node dropdown with a permanent left-rail
Replaces the /me/workflow-editor entry that dropped users straight into
Idempotent host-pass DataSeedContributor that materialises the three wizard
Foundation for the six host-admin /media surfaces.
Catalog entity for wizard-surfaced workflow presets.
Public-site MVC Razor pages were rendering hardcoded fixture data
DbMigrator hit E11000 on BlogPostsDataSeedContributor: the legacy
BlogPostsDataSeedContributor adds the initial public marketing blog catalog
PublicWebsite layout header override:
- Anonymous browse for /marketplace (paged listings)
- Pages/Sitemap.cshtml (route /sitemap.xml) — returns sitemaps.org urlset
- Pages/Legal/Detail.cshtml (route /legal/{slug}) — Markdig-rendered long
- Pages/Pricing/Index.cshtml (route /pricing) — Monthly/Yearly toggle (real
- Pages/Docs/Index.cshtml (route /docs) — section grid landing + sidebar
- Anonymous browse for /personas (paged catalog)
Pull landing data from IPublicLandingAppService stub + pricing teaser from
- ToRun.Web.Public.csproj — add Markdig 0.37.0 for markdown rendering
- Anonymous read for /legal/{slug} — Terms, Privacy, Cookie, Refund,
- Anonymous read for /faq
- Anonymous read for /docs (tree + by-slug)
- Anonymous read for /changelog (paged, PublishedAt DESC)
- Threads list (paged, optional CategoryKey tag match against ForumThread.Tags)
- Anonymous read surface for /blog (list + by-slug)
smoke-backend.py flagged a VERB_MISMATCH: Angular calls bare PUT
Backend half of Sprint 50-F: anonymous landing-page demo chat with per-IP
The dictionary `Dictionary<(DateTime Date, Guid?
Closes the only race window in KeywordSettlementJob that could
Continuation of task #67 (50-SS sweep).
Closes the 50-RRR foundation (commits A/B/C) by wiring the actual composite
Backfilled Toggle / Heading / New / Untitled / Empty / Load:Failed / JustNow / MinAgo / HourAgo / DayAgo across all 28 non-English locales (en.json already had the full set from Sprint 50-CCC-2).
Mirrors Sprint 50-LL (Post.CreatorProfileId) + Sprint 50-VV-2 (AdAuction
Pipeline now derives the capability key from input topology and uses the
Adapter contract for the multi-input composite path.
Foundation commit for the multi-image edit refactor (Option C — capability
Scope finding: the wire-up at ChatPipelineAppService.cs:1415 already calls
5 structured log additions so production failures surface their full context
Two-pronged fix for the bubble accumulating empty / unwanted sessions:
Closes the deferred -2 from 50-YY (where the mic-attach was carved out so the
Closes the cross-pipeline gap left by 50-AAA.
User reported both "Pick a source image" and "Pick a mask image"
Production log:
User reported both source-upload and mask-upload on /me/image Edit
Final library module slice.
User feedback: green "Upload complete." banner flickered in/out for
Sprint 50-CCCC tried to fix the upload-pipeline ObjectDisposedException
Root cause of the persistent "Upload failed" toast — even though files
User reported five separate issues:
User reported part_001.wav / part_002.wav deleted from library but
50-YYY left a regression: openAssetDetail only triggered text preload
User reported Documents tab empty after uploading .md / .txt files —
CDN public URL pattern was wrong for /me/library — content is per-user
Two bugs in one batch — both surfaced by a user test where 4 files were 'uploaded' per the green banner but the library page stayed at 0 items.
User: 'upload işlemleri için progress bar koy, uploading vs saniye yazsın hangisi upload oldu kullanıcı bilgilensin, böyle çok yavan duruyor.'
Two user-blocking issues from /me/library testing:
User reported library upload 400 'Only image/* MIME types are accepted on this endpoint.' after uploading a non-image file from the All tab.
User reported Message Citations admin 500 'Internal error'.
User reported library upload 'Bunny storage upload failed with HTTP 401'.
User reported /me/image Generate broken: "Image generation provider
User reported the Sprint 50-III panel-position refactor is STILL not
Root cause (Scenario B variant): `openSession()` calls `messages.set(mapped)`
User: "You don't have to use the same model for titles, these are
User reported panel kept opening detached from FAB despite Sprint
Sprint 50-CCC raised MaxOutputTokens 24 → 128 expecting "~100
User reported /me/activity-history empty despite hours of chat/image/voice/etc.
Two user-testing regressions found and fixed; the third concern
| # | Menu | Group | Angular path | Backend route | Status | Root cause |
User: "library dosya eklerken neden tek tek el ile eklemem gerekiyor,
User: "source da upload from pc var da mask ta neden yok, bizim
User-reported critical bug: every PC upload attempt threw
User reported repeatedly since yesterday that /me/notifications/
Recent-chats list in /agent/chat-pipeline had per-row Rename/Delete
User testing surfaced 5 concerns; 4 of them addressed in this commit, 1
Daily AsyncPeriodicBackgroundWorker (24h tick) sweeps Paid
Sprint 47-O created flat AdAuction entities without join FKs;
Daily ABP AsyncPeriodicBackgroundWorker (24h tick) sweeps pending
Sprint 50-TT was dispatched for this but went off-script (it did
User asked: library top size indicator, PC upload, plan-tier kota,
Extends the 50-RR pattern (VideoJobPollingWorker IAbpDistributedLock)
Sprint 50-P deferred PC upload because backend lacked a Bunny
50-AA aborted during the full sweep; this is the resumed batched
Sprint 50-U + 50-EE wired polling + refund flow but assumed single-pod
User direktifi (Sprint 50-E turn): bell + activity sağdaki menü ikonlarının
Backend unreachable at https://localhost:44304 during this sprint;
Sprint 47-H built a degraded compound index for Creator/Post feed
Sprint 47-H built a degraded (TenantId, CreationTime) scan for PersonaPost
Sprint 50-W BillingReconciliationJob detected holes (Paddle
Each modules/torun.*/angular/angular.json referenced
Sprint 50-CC tenant isolation tests revealed a pre-existing defect:
Sprint 47-A..47-P MT sweep added TenantId-prefixed compound indexes to every
Sprint 49 series completion.
Pre-launch R2 risk: Sprint 47-A..47-P swept 219+ entities to
Sprint 50-U surfaced gap: IBillingRecorder had only RecordAsync.
Plan §2.3 / R1 audit — Sprint 47-A...47-P swept 219+ entities to
Plan §2.5 backlog: propagated 22 ImageEdit pipeline keys (Sprint
Sprint 50-K caught one collision (subscription plan paddle config
Phase 1.5 §2.1 #3 — third background worker.
Phase 1.5 §2.1 #4 — fourth and final pre-launch background worker.
/api/app/run-input-upload returned 404 because the AppService was
Sprint 50-P landed the /me/image Edit tab UI + backend wiring but
Frontend (Angular | date pipe + Date constructor) handles UTC -> local
Phase 1.5 §2.1 — top-priority background worker.
Phase 1.5 §2.1 #2 — second background worker.
User reported "form doldurma" follow-up turn returned "Message failed.
Bug A: chat composer ignored Ctrl+V on clipboard images.
The Marketplace browse page (angular/src/app/marketplace/
Edit tab was missing the Model selection (Auto/Manual), Quality
User reported the Sprint 50-I default (bottom:96 right:24) overlaps
User testing /me/notifications hit 404 on
PaddleOptions.WalletTopUpPriceIds was a Dictionary<string,string>
User reported "paddle.js popup açması lazım, hata veriyor" when
Sprint 50-I addresses two UX regressions surfaced after Sprint 50-B + 50-E:
Admin saved a SubscriptionPlan's Paddle config; the first save worked.
Cost breakdown table on /me/billing was naively prefixing every
Sprint 49-B left two hard-coded /api/app/* URLs in chat-pipeline
- 50-B (InAppAssistant:Bubble:Dock/Undock/DragHint): added to en.json + all 28 other locales with native translations
User: 'memory sayfasında sil tıkladığımda js confirm çıkıyor biz neden leptonx e
Fix A (server CRITICAL bug):
User reported: SourceKind dropdown showed weird pink/blue blocks (broken
User feedback: the two floating bells (notification-bell + activity-center)
Sprint 47-I + 45-M pattern apply.
Sprint 46-A introduced cross-module IBillingRecorder + emitted BillingRecords
Sprint 46-F structural normalize completed lowercase→PascalCase but left 4 locales with Suite-stub English content.
Sprint 45-M + 47-A through 47-N pattern apply.
Sprint 47-M pattern apply.
Sprint 45-M + 47-K pattern apply.
Sprint 45-M + 47-A through 47-K pattern apply.
Sprint 45-M + 47-A through 47-H pattern apply.
Sprint 45-M + 47-A/B/C/D/E/F/G/O pattern apply.
Sprint 45-M + 47-G pattern apply.
Sprint 45-M + 47-A/B/C/D/E/F pattern apply.
Sprint 45-M + 47-A/B/C/D/E pattern apply.
Sprint 45-M + 47-A/B/C/E pattern apply.
Sprint 45-M + 47-A + 47-B pattern apply.
Sprint 45-M + 47-A/B/C pattern apply.
Sprint 45-M + 47-A/B pattern apply.
Sprint 45-M + 47-A pattern apply.
Sprint 45-M pattern apply: Suite did NOT emit IMultiTenant on Base classes despite IsMultiTenant: true in JSON.
Slice 56 — IModelRoutingService (ByokVault.Application + Contracts):