--- title: Changelog description: Notable product and API updates for Sume. --- This page gives short notes for users about Sume product releases. The newest release is first. This page is the source of truth for readers about what shipped. ## v1.0.3 — October 18, 2026 Genjutsu, Cinema Studio 4.0, and Studio fixes - Genjutsu adds Motion transfer and Objects swap, priced at $0.318, $0.681 and $1.632 per second - Studio Genjutsu tab gets a new model and layout; Motion library and Styles are removed - Face swap, Language swap, Caption and AI Influencer open on the same hero card - Cinema Studio 4.0, AI Influencer and Ads Studio run on signed-in surfaces at listed prices - Genjutsu submits without a prompt; the Studio wall no longer shows dead sandbox links ## v1.0.2 — October 18, 2026 Wan 3.0 routing, Home asset search, Open fixes - Wan 3.0 and Wan 3.0 Prime text-to-video use a new primary route, with fallback if it refuses - Customer prices for Wan do not change - Home search finds assets on the wall - In an Enterprise workspace, Add workspace and Max ask the Sume point of contact - Open: the signed-out Ask Sume sends you to sign in and keeps your message ## v1.0.1 — October 18, 2026 Balance bar and Enterprise early access - The profile menu shows one Balance bar for plan and credit - Early access admits Enterprise workspace members and accepted Enterprise invites ## v1.0.0 — October 17, 2026 Reliability fixes - Internal reliability and deploy fixes; no user-facing change ## v0.4.69 — October 2, 2026 Faster settle, hold fixes, shared thread jobs - A finished thread no longer shows a hold, and a stopped turn settles in about a minute - Holds show on /v1/usage - A generation create that fails after its row is written no longer strands a hold - Thread participants can read the thread's jobs, whoever created them - A text-to-speech result records its generation_config and speed ## v0.4.68 — October 1, 2026 Blog on a new store, stopped-turn billing, GPT-6.1 Sol default - Blog posts move to a database and blog images to media.sume.com; old image URLs redirect - A stopped turn bills every provider request it sent - GPT-6.1 Sol is the Formats API default when no model is given - The video router adds H3 Max Recast (h3-max-recast) - The lightbox opens on processing and failed assets; only the author can edit a chat message ## v0.4.67 — October 1, 2026 Turn reliability, Auto fallback, model marks - Turn close no longer waits on a slow rate card, and the reply saves before billing - A turn that is still working is not stopped - Auto falls through a hung DeepSeek host to the next model, even after a stream cut before the first token - The model pill and menu rows show the maker's mark - 632 more English blog posts ## v0.4.66 — October 1, 2026 Security updates, MCP and Formats improvements - Security: web framework update for critical advisories; MCP OAuth consent accepts only MCP-host tokens - Web writers act only in the workspace the page names - MCP jobs_wait returns finished results inline; a Format run lists only the tools its package declares - A cold start shows "Starting sandbox…"; provider credit exhaustion has its own error class - A Format run create answers 202 instead of 503; a killed worker no longer resubmits a paid provider stage ## v0.4.65 — October 1, 2026 GPT-6.1 Sol and GPT Image 2 pricing - GPT-6.1 Sol joins the catalog beside GPT-6 Sol, which stays - GPT Image 2 is priced from the provider's per-image table - A GPT Image 2 request with no size reserves the square bucket - More English trend search-query blog posts ## v0.4.64 — September 29, 2026 Faster Agents and Assets, 365 blog posts - About 1 MB less JavaScript on every Agents route - The Assets rail paints only what is visible, and the Assets wall fills a few rows per idle step - Typing in the home composer re-renders only the composer - Hovering a thread row starts its transcript read, so opening is faster - 365 English trend search-query blog posts with thumbnails ## v0.4.63 — September 29, 2026 More blog posts - 206 buyer search-query blog posts - 200 trend search-query blog posts with thumbnails - 12 existing posts rewritten ## v0.4.62 — September 29, 2026 Claude Sonnet 5.5, list-price billing, faster Assets - Claude Sonnet 5.5 replaces Sonnet 5 in the model picker - Turns and sidecar calls charge list price per request; this fixes the multi-call over-count - A plain ask reaches reference remix as a first-party skill; Canvas opens only on the thread that asked - Assets list reads are faster, and hidden tabs stop background polling - A workspace can be set to 10,000 concurrent generations; lists page past ten videos or avatars ## v0.4.61 — September 27, 2026 Act-first agent turns, 200 blog posts - Agent turns do the asked action first, with no catalog or notes reads before it - MCP creates skip the list and cost preflight - A pasted reference link is read with reference_ingest - Board opens and thread lists write and read less - 200 query-first blog posts in English and Korean ## v0.4.60 — September 27, 2026 100 more blog posts, category hubs, RSS - 100 more blog posts in English and Korean, each with a thumbnail - Blog category hubs at /blog/category/, and RSS at /blog/rss.xml - The Korean API reference adds the Format-run webhook redeliver row - The reference remix kit inks the spoken word in its speaker's colour ## v0.4.59 — September 26, 2026 100 new blog posts, media and web speedups - 100 new blog posts in English and Korean, each with a thumbnail - Faster media processing: frame stills, inspect and compose, and reference ingest - Marketing pages load faster, and the first Assets screen streams ## v0.4.58 — September 26, 2026 Reliability fixes - Internal reliability and deploy fixes; no user-facing change ## v0.4.57 — September 26, 2026 Korean blog, CDN-served landing, one model registry - Korean versions of all 45 blog posts at /ko-KR/blog - Anonymous visitors get the landing page and /ko-KR from the CDN - The sitemap and hreflang list a Korean URL only where a Korean page exists - One registry row per model id feeds the picker, billing and Usage - Faster video frame reads and usage metadata reads ## v0.4.56 — September 25, 2026 Blog guide series and thumbnails, agent and worker fixes - 30 answer-first guides on Formats, the developer platform, agents, Avatar 1.0, media tools and models - Every blog post gets a thumbnail - A finished turn no longer shows as failed when its settle request fails - A render waits out a short media-mirror failure instead of being discarded ## v0.4.55 — September 24, 2026 Blog comparisons, safer deploys, auth and billing fixes - Eight "Sume vs X" comparison posts with sourced, dated facts - Releases no longer cut live MCP sessions - Workspace and auth hardening: Agents routes act in the named workspace or refuse, MCP consent grants what it shows - Billing: held lots expire with their plan, and out-of-order subscription webhooks no longer roll a plan back - A GPT-6 Luna pick now runs GPT-6 Luna through Usage ## v0.4.54 — September 23, 2026 Billing resilience, media performance, model picker - Billing: expired credit holds, fee splits, auto top-ups and delayed payments are handled correctly - Auth keeps session tokens on their own origin, and team admin writes stay in their team - Faster media output, probe and upload - The picker orders models by company, retires GPT 5.6 and Opus 5, and routes GPT 5.6 Sol to GPT-6 Sol - The dashboard enters with no blank plane; Run input opens a real preview of a Formats run input ## v0.4.53 — September 23, 2026 Right-pane tiles, new models, billing fix - Right-pane tiles are on in production, with a Layout menu and split panes - GPT-6 Sol, GPT-6 Luna and Claude Opus 5.5 join the catalog - A paid Checkout is credited even after an earlier failed attempt ## v0.4.52 — September 22, 2026 Prompt metering, Usage labels, thread fixes - Billing meters the full agent prompt - Usage shows the locked 5.5% platform fee rate - Automatic thread titles are routed and priced as sume/auto - Edited branches no longer return during reconcile ## v0.4.51 — September 22, 2026 Thread Usage and attachment editing - Click-to-edit keeps attachments and restarts edited branches - The thread cost chip and Usage share one spend read ## v0.4.50 — September 22, 2026 Usage, pricing, media, and dashboard improvements - Thread cost chip reads from the ledger and updates live without a reload - /v1/usage reads thread, run and job spend - Listed SKU prices are list × 1.25, with a 5.5% platform fee on spend - MiniMax H3 Max lip-sync joins Fabric on avatar-image-to-video - Dashboard tabs open at the top of the content plane ## v0.4.49 — September 22, 2026 Faster Agents navigation - Send opens the thread route at once, and the cold shell shows the thread frame - Thoughts completed by a gap persist on replay ## v0.4.48 — September 21, 2026 Grok 4.7, Script run details, and Thought clock fixes - Grok 4.7 retires Grok 4.6 in Settings LLM; saved Grok 4.6 threads keep their identity - The Script run row names the calls inside, with counts - Settled tool rows keep their details and body on reload - The Thought clock no longer bridges gaps or reused rows ## v0.4.47 — September 21, 2026 Format grants, /videos inspector, instant lightbox - Owners can switch a Format's API availability on or off - A member's team key defers to the Format grant instead of returning 403 - A grantee key never bills the Format owner - /videos adds an inspector beside each clip - /images opens the lightbox instantly ## v0.4.46 — September 21, 2026 Music Router, in-thread speech edits, video thumbnails - Music 1.0 runs on Lyria 3.5 through a new Music Router (sume/music-auto) - Speech edits apply in the thread with no review step; text to speech is literal by default - Attached videos show as preview thumbnails in your message - Script run lets an agent call remote MCP tools programmatically - Hyperframes can reuse earlier media builds, so caption re-renders cost nothing ## v0.4.45 — September 18, 2026 Parse-first speech edit selection - Speech edits for source-bound text to speech select the target text before editing - Punctuation-only fragments stay inside the selected speech target - Speech approval no longer races with turn claims ## v0.4.44 — September 18, 2026 TTS speech edit and language warning - Warns before a voice and text language mismatch in TTS - Lets you confirm and retry after the warning - Accepts reviewed chat speech edits before TTS runs ## v0.4.43 — September 18, 2026 Tagalog, Soul 2, Gateway keys - Adds Higgsfield Soul, rebuilt on SOUL 2 - Retires the Image 1.0 model and uses the Higgsfield logo for Soul - Adds Tagalog voice support - Adds Vercel AI Gateway workspace keys (BYOK) with billing details in Usage - Canvas opens as a read-only Assets view; pinch zoom is more sensitive ## v0.4.42 — September 17, 2026 Faster first token, TTS sources, Format receipts - Cuts time to first token for agent turns - Fixes TTS source handling per workspace and thread - Shows thread authors by workspace member, never a raw account id - A failed Format run now ends as failed, not completed - Sweeps Format runs that stay stuck on running ## v0.4.41 — September 17, 2026 HyperFrames, MCP and OAuth fixes - Fixes HyperFrames clip timing, audio and media-type checks - MCP and OAuth endpoints return correct errors on unsupported content types - Bounds turn ingest so batches no longer cancel live jobs - Updates the Grok Build CLI to 1.0.35 - Canvas tiles load media correctly; empty right-rail tiles are flat and monochrome ## v0.4.40 — September 16, 2026 Automatic first-party skill updates - First-party skills and Format cards now update automatically after each release - Fixes Canvas rename and right-rail windowing - Fixes how a failed turn renders in a thread ## v0.4.39 — September 16, 2026 Recreate, Restyle and Canvas on production - Recreate and Restyle cards appear on production - The right rail shows Canvas and Moodboard - Fixes overlay identity, collision and backfill issues - TTS now uses accepted transcript sources - Improves attachment handling and talking-shot guidance ## v0.4.38 — September 16, 2026 Assets Generated hub and restored rail - Adds a Generated hub in Assets with an Overview and Library rail - Restores the Chat, Format, Assets and Moodboards left rail - Improves video inspect with bounded transfers and a byte fallback - Browser sessions end when idle and keep context per thread - Cleans up avatar prompts ## v0.4.37 — September 15, 2026 Assets quality, Moodboards, video inspect - Adds per-model Quality and Resolution controls in Assets, including Nano Banana 4K - Ships the full Moodboards update - Speeds up video inspect with a fetch-time probe - Removes the higgsfield-soul model from the image API routes ## v0.4.36 — September 15, 2026 Agents Browser on production - Agents Browser is live: start a session, navigate and end it - Adds Browser usage reporting and bounded sessions - Formats gets pinned chrome and initial tiles ## v0.4.35 — September 15, 2026 Archived threads, rail and Formats - Adds an Archived filter and Unarchive for threads - Widens the right rail - Formats with no art show an empty tinted face ## v0.4.34 — September 15, 2026 Agents web chrome - Fixes workspace chrome when sign-in is rate limited - Collapses the sidebar on the first send - Shares one chrome across project pages - Polishes Formats ## v0.4.33 — September 15, 2026 Reliability fixes - Internal reliability and deploy fixes; no user-facing change ## v0.4.32 — September 14, 2026 Gateway throughput, folders, timeline - Agent model gateway ranks providers by throughput - Fixes Project folder origins and hidden-origin hints - Timeline trims before changing fps - Fixes attachment URL copying and stream gaps - Fixes Keys and BYOK screens ## v0.4.31 — September 14, 2026 Caption tools and agent speed - Caption tools search plugins and keep font identity - Hides Meta Ads tools until they are configured - Improves perceived speed of agent turns ## v0.4.30 — September 14, 2026 Stream retries, MCP and thread media - Retries dropped model streams and cleans up cut-off stream copy - MCP now fails closed and builds a tool manifest per turn - Thread media loads in a better order - Playing video shows its stage; timeline renders get a poster frame ## v0.4.29 — September 14, 2026 Markdown bubbles and steer fixes - User messages render GitHub-style Markdown - Steer state survives remounts and shows a stopped label - Timeline labels low-fidelity speech-to-text WAV audio - Shows Attaching only while the workspace write runs ## v0.4.28 — September 13, 2026 Folder paging, Usage CSV, steer - Pages Project folder threads - Usage CSV export has 15 columns in a fal-compatible format - Steer shows on first paint - Enter dequeues or steers a queued message - Improves time to first token ## v0.4.27 — September 13, 2026 Usage CSV and Dashboard Scheduled - Adds a three-row rail and a Scheduled view in Dashboard - Usage CSV matches the fal-compatible format - Usage CSV adds a raw billing formula and pricing_units_json columns - Adds the Meta Ads mark ## v0.4.26 — September 13, 2026 Faster Usage and billing balance - Usage page loads faster, with fewer database waits - Billing balance loads faster - Usage totals query runs faster - Assets rail, notifications and thread list load faster ## v0.4.25 — September 13, 2026 Usage, Integrations and Meta callback - Usage 30-day totals are computed in one query - Integrations shows Slack, Meta and Google as Coming soon - Fixes the Meta OAuth callback - Adds privacy deletion instructions - Fixes Slack teammate join ## v0.4.24 — September 12, 2026 Release fix - Completes the v0.4.23 release - Includes UI fixes made after v0.4.23 ## v0.4.23 — September 12, 2026 Moodboards, Filters and Integrations catalog - Closes a Moodboards page rollout gap - Adds Filters and social embed updates - Updates the Integrations catalog - Disabled first-party packages are hidden from discovery and blocked from running ## v0.4.22 — September 11, 2026 Auto retries and Slack replies - Auto hops to another model when the first turn fails - Slack follow-ups reply in their own thread ## v0.4.21 — September 11, 2026 Fonts catalog and Assets rail - Search fonts and stage an exact revision with coverage and license checks - The right rail shows Generated assets, including failed and completed states ## v0.4.20 — September 10, 2026 Auto compose, voice and Omni video - Improves Auto compose - Voice requests accept raw provider voices - Omni video works with Auto - DeepSeek models move to general availability ## v0.4.19 — September 10, 2026 Reliability fixes - Internal reliability and deploy fixes; no user-facing change ## v0.4.18 — September 9, 2026 Agent recovery, DeepSeek and captions - Fixes agent stream and recovery issues - Adds DeepSeek model support and Auto routing - Improves captions and video inspect - Adds Meta read, import and remake guidance - Changes first-party skill lifecycle and gates integrations ## v0.4.17 — September 9, 2026 Startup readiness - API waits until it is ready before it accepts MCP traffic - A cold start no longer returns 429 rate_limit_unavailable - Keeps sign-in cookies on prefixed URLs ## v0.4.16 — September 9, 2026 MCP fail-soft - One MCP server startup miss no longer stops a turn - The turn continues after MCP startup misses ## v0.4.15 — September 9, 2026 Attachment fix and GPT Image 2.5 Flare xhigh - Attachments no longer fail on the image pre-check step - GPT Image 2.5 Flare adds xhigh and max quality - Display name saves correctly - Sidebar Show more works past 225 items - Rail hover menu is shorter ## v0.4.14 — September 9, 2026 GPT Image 2.5 Flare - New model: GPT Image 2.5 Flare - Image quality is set only when the model catalog offers it - Retiring Image 1.0 and Video 1.0 inspect aliases route to Router Auto - MCP broker retries on DNS errors - Compact tool bundles start collapsed ## v0.4.13 — September 8, 2026 Thought rows and Assets rail - Thought rows show a carried thinking tail once and hide replays - Assets rail wakes for avatar and audio output - Assets rail paints bursts of results progressively ## v0.4.12 — September 8, 2026 Agents chrome, Korean TTS, Format receipts - Agents: no duplicate Thought rows, no Reconnecting row, Planning in the compact slot - Agents: Generating, Running, and Ran rows with a count line - Assets filter: All, Generated, In Chat (default All) - Korean TTS requires Hangul text - Video frame aspect and timeline fps now match the source - Format request JSON download; H3 Max 1080p in the catalog ## v0.4.11 — September 8, 2026 Live tool chrome and API key spend - Spend keeps the API key that started the request - Running wrap-up header uses the -ing form; provider errors keep the turn history - Exploring bundle children align to the header - Jobs tab scanability polish - Agents prefer the jobs_wait MCP tool over the terminal ## v0.4.10 — September 8, 2026 Agents chrome polish - Working clock and fold rhythm refined - Native Grok Imagine block - Tool and terminal output in a code box - Profile chip opens settings - Exploring collapses per tool chunk ## v0.4.9 — September 8, 2026 Agents chrome and MCP admission fixes - Planning shows on a quiet stream instead of Waiting for connection - MCP admission echoes the JSON-RPC id and uses a shorter wait budget - Working for timer stays live after a remount - Live tool titles stay through settle - Sidebar sorts chats by last activity ## v0.4.8 — September 7, 2026 GPT 5.6 Sol default, Rules page - GPT 5.6 Sol is the default Agents model - Auto routes to Gemini with a same-turn Sol fallback - Workspace Rules page redesign, with an empty state for zero rules - Formats: Archive menu and a heart and fork stack on tiles - Chat shows live personal display names ## v0.4.7 — September 7, 2026 Workspace Rules and Creator member UI - Workspace Rules are added to /workspace/AGENTS.md for agents - Admin-only member controls are hidden from Creators - /v1/videos catalog docs updated ## v0.4.6 — September 7, 2026 Crawl timeout, images_review removed - crawl_scrape now fails with a clear timeout - The images_review tool is removed ## v0.4.5 — September 6, 2026 HyperFrames checks and Formats API docs - HyperFrames check reports added - Failed polls stop instead of repeating - Dashboard hides the concurrency plan table - Formats API docs updated ## v0.4.4 — September 6, 2026 Fable 5 aliases move to Fable 5.1 - Fable 5 retires; its aliases route to Fable 5.1 - Past Fable 5 usage keeps its saved rates - Public video-frames guide added to the docs ## v0.4.3 — September 6, 2026 Creator role messages, Claude resume fix - Clearer message when a Creator role is unavailable versus a missing team - Foreign reasoning is scrubbed before a Claude session resumes ## v0.4.2 — September 6, 2026 Fable 5.1 and stream reliability - New model: Fable 5.1 - Agent streams are bounded - Recoverable failures give clearer diagnostics ## v0.4.1 — September 6, 2026 Format harvest fix and billing ledger - Format runs with an accepted output stay successful after a harvest error - Billing ledger: one rate table, price quotes, settlement, spend summaries - Hung provider calls are cut off by a watchdog - Fixed the gpt-image-2 price bucket - MCP check-call shapes fixed ## v0.4.0 — September 6, 2026 Format host and skill updates - Format phase and continue support - Terminal webhook and receipt updates - Bounds on API and MCP work - Updated crawl, media, and bundled skills - Native Grok mode and medium effort ## v0.3.98 — September 5, 2026 OpenAPI and SDK sync, GPT-6 Astra - OpenAPI snapshot and SDK client in sync - GPT-6 Astra added to the LLM catalog - Codebase refactor and documentation overhaul ## v0.3.97 — September 5, 2026 Settings model wins - Unset Settings use the live default model on every thread - In a shared thread, each turn uses the sender's Settings model - Social ingest, voice clone, and caption hard-cut updates ## v0.3.96 — September 4, 2026 Thinking row, Settings model picker, Format run model - Thinking row opens and carries thoughts across batches - Settings has an Agents model picker; composer model and quality chips are removed - Format run create accepts an optional LLM model; omitted stays Sol - Account menu scrolls the org list ## v0.3.95 — September 4, 2026 Installable app and product mentions - Sume can be installed as an app from the menu - The @ mention product list refreshes when a product is saved ## v0.3.94 — September 4, 2026 Video and audio tools, aspect ratios - New media tools: video_inspect, video_trim, audio_detach - Image models receive native aspect ratios - Tool rows paint on resume and wake, with no replay on reattach ## v0.3.93 — September 3, 2026 Member role is view-only, Brands add product - Member is view-only; running actions and writing threads need Creator or above - Format create: hard copy versus soft copy - Brands: add a product by link scrape or by hand - Landing Pricing links open the Enterprise booking modal - Steering keeps the interrupted turn's band ## v0.3.92 — September 2, 2026 Image Upscale uses Topaz generative - Image Upscale 1.0 runs on Topaz generative; the public id and body stay the same - Image Upscale reserve is $0.16 per image - Video upscale is unchanged ## v0.3.91 — September 2, 2026 Goal chrome removed, image clip render fix - Goal set chrome is removed from Agents - Still-image clips are fetched during composition render - Render fetch failures are classified ## v0.3.90 — September 2, 2026 Video frames API and usage ledger - New /v1/video-frames and /v1/hyperframes-previews endpoints with MCP tools - LLM usage ledger tracks three money values and settles usage ## v0.3.89 — September 2, 2026 Agent billing fix, HTML as source of truth - Every model request in a turn is billed - Cache usage is split into separate buckets - HTML is the stored composition document; Canva saves serialized HTML ## v0.3.88 — September 2, 2026 Caption render hardening - Caption render retiming uses a shared root locator - Three new non-blocking caption check warnings - Additive response headers from the render service ## v0.3.87 — September 1, 2026 Webhook delivery for non-owner members - A non-owner member's Format run posts the stored webhook - return_format_output uses the thread workspace - New response fields: scope, cited_jobs_read, webhook_destination ## v0.3.86 — September 1, 2026 Render check fix - A stale render image no longer counts as a verdict on the composition ## v0.3.85 — September 1, 2026 Team wallet default, media verification - New teams get the pooled wallet on by default - Cited media is verified by the producing job id ## v0.3.84 — September 1, 2026 Formats API chrome, media claim fix - Formats API chrome shows only on the API-fired turn - Truncated ledger scans no longer claim media was never produced ## v0.3.83 — September 1, 2026 Agents chrome and webhook fixes - The idle send glyph stays while a turn streams - Connecting chrome clears once tools run - A webhook URL is enough to send return_format_output ## v0.3.82 — September 1, 2026 Caption stroke guard, design brief - A compose render refuses to drop an authored text stroke - Editor drag and nudge no longer flatten a text stroke - Public /design.md, /design.css, and an Agents design skill ## v0.3.81 — September 1, 2026 Caption CSS passthrough, design brief - Author caption CSS passes through to the render - The caption check no longer rejects its own recipes - Receipts include generated media from the same workspace - Public /design.md brief for the Agents tab ## v0.3.80 — September 1, 2026 Caption fixes - Composition caption rows save with a valid style ## v0.3.79 — September 1, 2026 Jobs and Format run reliability - Interactive chat reads jobs from the same workspace - A job stopped by an operator shows a clear status - Format run create has its own timeout and a truthful timeout message - Leftover queue claims expire ## v0.3.78 — September 1, 2026 Queue controls and caption bake - Queue actions show without hover - Compact and touch layouts get a queue control - Draining no longer eats an in-progress queued edit - Caption bake fails when style attributes are truncated ## v0.3.77 — September 1, 2026 Speech boundary fix - Speech boundaries resync and persist after a replay ## v0.3.76 — September 1, 2026 Image tiles, new commands, caption HTML - Every generated image tile stays after a replay - New /brief-format, /face-swap, and /product-swap - Caption presets store HTML ## v0.3.75 — September 1, 2026 Formats API profile, caption chrome - Formats API profile and webhook updates - Caption hold and chrome fixes - Cursor picker and Goal updates in Agents ## v0.3.74 — September 1, 2026 Kling Motion Control - Kling Motion Control in the @ catalog - Copy and Comment actions - MiniMax H3 Max added ## v0.3.73 — September 1, 2026 Job queue, Formats generate, model picker - Jobs queue past full concurrency and free stuck processing seats - Generate from Format chat - Formats API records the actor name - Agents model picker, Working fold, and Worked for margin updates - Reference remix URL binds correctly ## v0.3.72 — August 31, 2026 Video and Formats polish - Video Play button fill; wait-verb labels removed - Formats: shared like and a new nav glyph - Tool-row arrow updated ## v0.3.71 — August 31, 2026 Agents notes and chat folding - Tools inserted mid-chat fold into work groups - Each thought shows its own time span - Per-thread Notes tab - Caption fix for captions that straddle a render window - Pricing nav opens a booking modal ## v0.3.70 — August 31, 2026 Workspace state reliability - Large workspaces save durable state more reliably - Captions drop finished cues from time-slice render shards ## v0.3.69 — August 31, 2026 Shortform tools, fold polish - Shortform tool jobs_wait now waits until the job is terminal - products_mirror_image accepts a batch of URLs - images_review adds visual QA for generated images - Folded work waits until tool arguments are parsed - Workfold rows get more spacing and one type scale ## v0.3.68 — August 31, 2026 Composer, tab titles, shortform compose - Composer shows Stop and Update controls - Tab titles are updated - Lightbox is available through MCP - Shortform compose adds check_only and caption_burn - Public share pages get updated chrome ## v0.3.67 — August 31, 2026 Storyboard, Formats access, Thinking stream - Storyboard ships - Formats Access lists members and sets write grants - Streaming shows Thinking and Explored rows - Landing page shows a sample user turn ## v0.3.66 — August 30, 2026 Format invites, remix from scenes, spend breakdown - Formats get invite chrome and workspace grant invites - Remix can author a Format from scenes - Assets show a Heart and Copy public URL on hover - Usage and Billing break spend down by category - Cartesia Sonic 3.6 is the default voice - A stale tab shows an Update pill ## v0.3.65 — August 29, 2026 Remix without a key, partner playgrounds - Remix works in chat and Formats without a playground key - Partner Format playgrounds run the partner's own model - Partner brand intake works when brand data is missing - TikTok research rows include cover_url - The thinking fold persists, so a reload matches the stream - The project picker creates folders ## v0.3.64 — August 28, 2026 Partner publishing, remix formats - Partner Formats create and publish live, with a 402 prompt when funds run short - Partner Formats connect and publish to TikTok and Instagram - Remix analysis is turned into a Format before saving - The @ and / menus are split; /trending-research is added ## v0.3.63 — August 28, 2026 Inbox deep link fix - Inbox comment deep links open once ## v0.3.62 — August 28, 2026 Video analysis keyframes, Format chrome - Video analyses capture at least one keyframe per second of each scene - Format three-pane chrome sits on one baseline ## v0.3.61 — August 28, 2026 Phosphor icons, chrome alignment - Product icons switch to Phosphor - The chrome title band aligns with the sidebar nav row - Footer menus open in a portal - A paused chat that resumes empty restarts cold on the next send and on fork ## v0.3.60 — August 28, 2026 Resume failure handling - A resume that disconnects with zero tokens now fails the turn - The turn shows: Couldn't finish this turn. Please try again. - The next click on the same thread starts a fresh session ## v0.3.59 — August 28, 2026 Fork toast, receipt fix - Fork shows a toast when the source thread is still running - A fork cuts settled turns into a chat-only copy ## v0.3.58 — August 28, 2026 Gemini Omni Flash 1.1, Seedance audio, lightbox video - Gemini Omni Flash 1.1 joins the video router: text-to-video, image-to-video, reference-to-video, edit - Seedance prefers native audio per request - Chat video opens in the same lightbox as images - Agents Dashboard click opens workspace settings - Streaming tool arguments parse correctly in the fold, Flow, and inspector - Fork 502 errors are fixed ## v0.3.57 — August 27, 2026 Caption render fixes - Caption composition no longer fails with a TypeError - Caption encode failures show the real error, not a generic one - Caption boxes rematch on first compose, so the bake matches the editor preview - Workspace picker org logos fit the row height - The header toggle and new chat button hide while the sidebar is open ## v0.3.56 — August 27, 2026 Caption boxes, compact thread menu, scroll follow - Default caption boxes hug the text and sit in the bottom safe band - Sidebar hide: the header holds a toggle and new chat - Thread menu is compact; Pin and Archive leave the bar - Overlapping zombie turns stop - Caption render runs in one pass - The stream follows only when the view is flush at the bottom ## v0.3.55 — August 27, 2026 Compact thread menu, scroll follow - Sidebar hide: the header holds a toggle and new chat - Thread menu is compact; Pin and Archive leave the bar - Overlapping zombie turns stop - Caption render runs in one pass - The stream follows only when the view is flush at the bottom ## v0.3.54 — August 27, 2026 Thread flyout, folder rename - Thread flyout: Move opens sideways, with More, Pin, and Archive - Folders rename inline - New folders pin at the top - Jobs and the job index are scoped to the thread ## v0.3.53 — August 27, 2026 Reading line, image lightbox - The streaming reading line sits above the composer - The new-folder draft is kept - Chat image previews open in the Assets lightbox ## v0.3.52 — August 27, 2026 MiniMax H3 Max, Formats org runs - MiniMax H3 Max is generally available for text-to-video and image-to-video - Members can read, cancel, and re-dispatch Format runs in an org workspace - The sticky user message in Agents chat is one float plate, flush to the top - Wallet credit-shortfall turns show plain in chat ## v0.3.51 — August 26, 2026 Format org webhooks, Kling 3.0 Motion Control - Format completion webhooks fire for runs in org workspaces - Kling 3.0 Motion Control is added - Shelf Format cards prefill ## v0.3.50 — August 26, 2026 Light Agents chrome, follow-up fix - Agents chrome gets light-theme tokens - Queued follow-ups run when the active job is orphaned ## v0.3.49 — August 26, 2026 Fork, composer, Platform nav, Formats on GPT 5.6 Sol - Fork a chat from any assistant reply - Composer keeps an opaque background when folded; extra air above it removed - Show more now collapses long messages to about 3 lines, not 6 - Platform link added to the top nav on /platform - Formats API runs on GPT 5.6 Sol - Thread menu opens in front of other content ## v0.3.48 — August 26, 2026 Agents chrome, Assets lightbox, hairlines - Agents chrome and scroll-stick behaviour refined - Assets lightbox shows the first frame for videos without a poster - Title and rail hairlines cleaned up - Folder and project captions updated ## v0.3.47 — August 26, 2026 Folder drag and drop, Usage list price - Drag the full folder row to move it; rename in place - Folder icon shows on hover - Usage shows list price - Folder view membership saves correctly - MCP tools show their real names in the agent tool list ## v0.3.46 — August 26, 2026 Formats edit, vanity GET, folder-drop hint - Formats edit write contract added - Contents GET accepts the vanity address - Pinned sidebar items align correctly - Folder drop shows a hint where the item lands - Agents can invent a workspace voice, then run text-to-speech with it ## v0.3.45 — August 25, 2026 Voices, caption editor, custom audio player - Voices: create flow split, bookmark and more actions added - Caption style editor shows boxes on the canvas - Hangul caption presets get stable slugs - Korean captions call to action updated - Rail and lightbox use a custom audio player - Queued follow-ups drain correctly ## v0.3.44 — August 25, 2026 Voices library, caption style creator, compositions - Voices library with list view, generate-first create, instant voice cloning and @ mentions - Caption style creator shows a preview first - Video Editor chrome sits on the composition stage - Thread compositions persist across reloads - MCP OAuth consent binds to the registered redirect URIs ## v0.3.43 — August 25, 2026 Avatars library, comment quotes - Avatars library with @ mentions - Comment quotes show a label and short text, no image - Source TTS and BGM stay after you apply a composition - Org concurrency cap is labeled as a workspace cap ## v0.3.42 — August 25, 2026 Webhook test and redeliver, caption styles - Send a test webhook from the console - Redeliver any single webhook call - Caption styles editor added - Right-rail Assets is the default tab - Lightbox shows model and prompt - Folders support drag and drop ## v0.3.41 — August 25, 2026 Folders, Sol recommended, custom image sizes - Folder sidebar for chats - Sol marked as recommended in the model picker - gpt-image-2 accepts custom pixel sizes - Org wallet works through MCP - Team vanity addresses supported - Home prompt and concurrency updates ## v0.3.40 — August 25, 2026 GPT 5.6 picker, Videos picker - GPT 5.6 added to the model picker - Videos picker added - Tool calls run in order - Minimized model-name pill in the composer - Sidebar Home shows a folder header ## v0.3.39 — August 24, 2026 Shared chat Working state, Formats bulk queue - Team members see the Working state in shared chats - Member write access restored in shared chats - Formats bulk run queue added - Sidebar syncs streaming status ## v0.3.38 — August 24, 2026 Team shared chats - Chats in a team workspace are shared with all members - This-Thread assets are visible to non-owners - Creators and above can write; Members read only - Wan 3.0 pass-through with clearer error text ## v0.3.37 — August 24, 2026 Formats output, webhook retry, caption presets - return_format_output added for Formats - Format webhooks honor 429 Retry-After - Fork clones the workspace - Caption presets support @ and clickable previews - Playable BGM library - Usage shows the amount actually billed ## v0.3.36 — August 23, 2026 Worked fold, HyperFrames captions, Usage landing - Working to Worked fold is steadier; Recents and sidebar pulses refined - Caption presets and HyperFrames captions from an attached video - Contextual BGM catalog - Dashboard opens on Usage; Overview counts agent LLM usage - Avatar-video jobs auto-retry after a content policy rejection - Formats Requests restyled; Playground output sizes to media ## v0.3.35 — August 22, 2026 Caption reliability, frontend speed - Captions name a missing sandbox asset instead of reporting a timeout - Queued caption jobs no longer fail with a bare HTTP 500 - Video editor panel and analytics load lazily - Assets API loads faster - Uppercase kicker labels retired ## v0.3.34 — August 21, 2026 Chat markdown reader, TTS Router, Hangul captions - Chat renders HTML, footnotes, math, Mermaid, GFM alerts, task lists and table copy - TTS Router with a Cartesia catalog; TTS 1.0 engine picker removed - Hangul captions with editorial emphasis and licensed Korean fonts - Usage page layout updated - Format repo pane shows dates and copies SHAs ## v0.3.33 — August 21, 2026 Format and Agents fixes - Format creation timeout fixed - Format tab updated - Assets rail fix - Thread title fix - Steer button flash removed ## v0.3.32 — August 21, 2026 Format handle addresses - Formats load by handle and slug address ## v0.3.31 — August 21, 2026 Reliability fixes - Internal reliability and deploy fixes; no user-facing change ## v0.3.30 — August 20, 2026 Hangul TTS, team Formats - Hangul text-to-speech - Team Formats resolve for workspace members - Agents chat stays healthy when skill loading fails ## v0.3.29 — August 19, 2026 Sidebar folders, generating tiles, landing - Sidebar thread folders - Generating tiles keep the reference image ratio - Landing page adds architecture and model cloud sections - Platform pages also serve from platform.sume.com ## v0.3.28 — August 18, 2026 Format chat, pricing margin - Format chat binding and parity, with an Open in chat button - Format run input surface and download - Tag chrome and contract block chips - Standard billable margin set to 1.25 over provider cost ## v0.3.27 — August 18, 2026 Korean-ad captions, jobs and MCP retries, faster Agents rail - Korean-ad caption style added to the captions API - Failed submits and terminal-failed jobs are no longer treated as retry cues - Agents sidebar and rail load faster - In-flight image and video jobs show in the Assets rail - Worked-for fold starts collapsed on history load ## v0.3.26 — August 17, 2026 Faster media fetch, Agents editor, save as format - Media inputs fetch in parallel, one lane per input - A stalled input fetch is bounded by rate - Save as format added - Agents editor, platform landing and TwelveLabs remix added - Skill reads no longer show as Shell - Composer pill snaps to the card ## v0.3.25 — August 17, 2026 Reliability fixes - Balance polling makes fewer upstream calls - Non-GET page probes return 405 ## v0.3.24 — August 14, 2026 Landing page updates - Landing composer Send opens the Book a call page - Pricing cards drag horizontally on all screen sizes - New headline: "Video Agent API built for Automation" ## v0.3.23 — August 13, 2026 Book a call on marketing pages - Public marketing calls to action now point to Book a call - Dashboard loading skeletons match the final layout - Artifact limit set to 30 minutes at 1080p ## v0.3.22 — August 12, 2026 Webhook signing secrets, Agents notify setting - Reveal your webhook signing secret from the dashboard - Rotate signing secrets per scope; old and new secrets both sign for 24 hours - Webhook delivery log and secret fingerprint added - Notify toggle moves to Settings - Agents Markdown replies read better: spacing and list indents fixed ## v0.3.21 — August 12, 2026 API key spend reporting - Per-key Monthly spend now shows spend that no key can claim - Format runs started by API generate through the agent sandbox, so this spend had no key owner - Before, this spend showed as $0.00 ## v0.3.20 — August 12, 2026 Jobs API, MCP and Agents trail fixes - `GET /v1/jobs` is filterable and pageable, with cursor pagination - Job 409 errors name the conflicting job and expose `idempotency_key` - MCP: `generate_image` is async, `jobs_result` takes `job_ids`, and `jobs_wait` is retuned - MCP: `playbooks_*` tools removed - Agents: tool trail detail stays visible, and long turns keep full tool history - Pinned sidebar sorts by recent chat activity - TTS maximum duration rises from 600 to 1200 seconds - Timeline accepts `audio.source_in` for section extracts without re-rendering ## v0.3.19 — August 10, 2026 Format spend cap defaults - Omit `generation_spend_cap_usd` and the platform default is $400 - An explicit per-run cap is honored up to $500 - Longer unattended Format runs no longer stop on the old default cap ## v0.3.18 — August 10, 2026 Public share polish, jobs scope, turn hold - Public thread share pages match the spectator chrome, with owner attribution - Shared threads open the assets rail by default and show only that thread's assets - `jobs_*` tools are scoped to the bound Agents thread - Agents hold turns that end with running jobs or without a deliverable - `generation_spend_cap_usd` is now optional ## v0.3.17 — August 10, 2026 Agents public share links - Share a thread by public link at `/{handle}/p/{threadId}` - Archived threads stop being shared - Enterprise teams show the correct plan label in billing summary ## v0.3.16 — August 10, 2026 Agents billing pool and Working timer - Agents read the unified credit pool correctly in billing summary - Fixes false "Agent $0" refusals - "Working for Ns" starts from the server time, so the timer is accurate ## v0.3.15 — August 10, 2026 Agents reliability and Working chrome - Agents hold the settle step while generation jobs are still running - Assistant text is saved before settle; streamed text no longer glues together - Working timer is monotonic and shows plain "Working" until the start time is known - Skills use `@handle/slug` ownership ## v0.3.14 — August 10, 2026 Custom Formats every turn, LLM modes, live capacity - Agents install every custom Format of the workspace on each turn - Agent LLM modes are now `luna` and `terra`, both on the Fast tier - Paid generation waves are sized from live capacity, not the plan ladder ## v0.3.13 — August 10, 2026 Enterprise concurrency floor - Enterprise concurrency floor is 20 ## v0.3.12 — August 10, 2026 Team invite and handle fixes - Invites to a full team return a clear `409 team_full` error - Admin-created teams get a workspace handle, so the switcher no longer shows "No such workspace" - Faster fixes for org and invite failures ## v0.3.11 — August 9, 2026 Partner Best practices example matches the real run - The Example create tab is now the same request as the documented 30 second live cut - Full `output_schema` is shown inline - English and Korean pages updated together ## v0.3.10 — August 9, 2026 Partner Best practices docs: create rules - `output_schema` is the full object with `name`, `strict` and `schema` - `generation_spend_cap_usd` must be non-zero and above the estimate - Script length is the only control of video duration - Check that the returned script matches your segments to catch truncated reads - Adds a measured create and clip-retry example ## v0.3.9 — August 9, 2026 MCP submit metering, live commerce talk lock - MCP charges one write per submit; polls are billed on read - Live commerce Format keeps talk as one clip per segment and runs every phase - Clip-retry docs lead with per-scene feedback in the instruction ## v0.3.8 — August 9, 2026 Most-used Formats first - The Format a workspace uses most is pinned first in My Formats - The same ranking applies to browse on `/agents` home - Update banner is redesigned; the call to action is the only control - Each talk slot in live commerce binds to its own voice-over segment ## v0.3.7 — August 9, 2026 Update banner, concurrency chart, banner card - Logged-in pages show a sticky "You're on an older version" banner - Concurrency chart defaults to the last hour - Provider 422 errors show real detail, including `input_media_unreachable` - `@` menu skills show the full `{author}/{slug}` handle - Team top-ups go to team billing, not the personal subscription - Live commerce banner is one padded white card ## v0.3.6 — August 9, 2026 Steer gesture, yearly plans, Current badge - Steer is one smooth gesture in Agents - Yearly billing for Pro, Startup and Scale plans goes live - Concurrency page hides a misleading "Current" badge for enterprise and team workspaces - Docs: partner clip-retry receipts and cost guidance ## v0.3.5 — August 9, 2026 Pricing refresh, steer restore, banner layout - Subscription plan ladder and self-serve concurrency ladder refreshed - Annual discount shown in the pricing UI - Enter queues and arrow steers again; steers land at once - New users default to "Run without asking" for spend approval - Org invite identity is resolved before membership is created - Live commerce banners keep content in the upper half ## v0.3.4 — August 9, 2026 Product image recovery, invite landing - Live commerce recovers product images when the product page blocks the fetch - Live commerce refuses to generate when there are no product images - Team invite accept page checks sign-in first and shows a legal footer - Landing nav has Products and Resources menus, a green status pill and a new footer ## v0.3.3 — August 9, 2026 Invite accept policy, @ menu, landing - Team invites expire after 14 days - Wrong-account and unverified sign-ins are blocked on accept - Consumed invite links show neutral copy; Resend creates a new link and ends the old one - `@` menu is shorter and opens below the composer on Agents home - Landing shows showcase chats and a new model order ## v0.3.2 — August 9, 2026 TTS in OpenAPI, onboarding, @ menu skills - TTS 1.0 is callable from OpenAPI - Receipts include `billable_amount_usd_micros` - Each succeeded scene returns its own `scene.video` - Format runs with an empty body are refused - `@` menu lists workspace skills, then Sume skills, then liked media - Onboarding survey answers persist; unfinished first-run resumes - Landing Agents chrome shows recent chats ## v0.3.1 — August 9, 2026 Agents settle fixes, landing QA - Fixes duplicate text with multiple Your Move prompts - Fixes truncated streamed replies after settle - Landing shows an Agents Home mock, wallpapers and a recreate clip ## v0.3.0 — August 9, 2026 Steering, new plan ladder - Steering and Your Move run in Agents - Fixes the Working and Worked label blink - New plans: $40, $120 and $400, with concurrency 1, 2, 4 and 10 - Existing subscribers keep their current plans ## v0.2.72 — August 5, 2026 Format input and output docs, faster pages - Format API docs cover `input` and `output_schema` in depth, with a structured output guide - Partner Live Commerce docs add FAQ, curl case study and scene retry - STT returns correct `words[]` and sentence segments - `stt_create` is available on MCP - Webhooks sign per recipient - Pages load faster: images, videos, brands, liked media and Formats ## v0.2.71 — August 5, 2026 Skills layers, API scale, workspace switch - Skills gain L0 and L1 layers and `sume/{slug}` Formats - API drains gracefully and applies plan-tier request limits - Workspace switch feels faster, with prefetch - Format OpenAPI contract is machine-readable - Format schema, progress and pagination are more accurate ## v0.2.70 — August 5, 2026 MCP generate tools, Formats by workspace - MCP: `generate_image` and `generate_video` replace `images_create` and `videos_create` - Format list follows the workspace of the API key - Format vanity URLs use the workspace handle - Format runs continue with `previous_run_id` - Unified credits in billing - Office documents attach in Agents - Agents stream more smoothly ## v0.2.69 — August 4, 2026 Studio Agent on GPT 5.6 - Studio Agent runs on GPT 5.6 - Agents and dashboard share one chrome - Waiting band fixes in Agents ## v0.2.68 — August 4, 2026 OpenRouter-shaped images and videos API - New OpenRouter-shaped `/v1/images` and `/v1/videos` endpoints - Matching MCP tools - Studio Agent sandboxes pause and resume - Automations tools bind to the right workspace - Faster page navigation ## v0.2.67 — August 3, 2026 Run webhooks and cancel - Run webhooks are no longer sent on `canceled` - Top-level `webhook_url`, `callback_url` and `mode` are accepted; a conflict returns 400 - Docs show Live delivery in English and Korean - Video Router docs are shorter ## v0.2.66 — August 3, 2026 Seedance restored on Video Router - `seedance-2-mini`, `seedance-2` and `seedance-2-fast` return to Video Router - Contract, catalog, OpenAPI, SDK and docs match - Team Formats show the team handle in lists and vanity URLs - Korean quick start is localized ## v0.2.65 — August 3, 2026 Team Format vanity, billing top-ups - Team Format list and vanity URLs show the team handle - Docs: team Formats are reachable over the API - Billing top-up credit packs - Format run webhooks docs are live - Live Commerce stall and timeline fixes ## v0.2.64 — August 3, 2026 Partner Best practices variants - Best practices show three Format call variants - Docs lead with Quick start, Sume basics and Best practices - Landing status indicator matches docs - Live commerce B-roll uses the speed preset at 720p ## v0.2.63 — August 2, 2026 Korean docs, 720p live commerce - Korean docs coverage rises from 17 to 61 of 62 pages - Korean AI-translation notice added - Live commerce uses 720p as the minimum - API call examples expand in The Basics and Formats docs ## v0.2.62 — August 2, 2026 Dashboard IA, Format webhooks, SDK 0.2.0 - Dashboard has Account, Workspace and Admin sections, with Team Usage and per-member spend - Format, Action and Completion run webhooks, plus status and result endpoints - Image attachments on Format runs and Agent Completions - Team Formats require a team API key - `@sume-com/sdk` 0.2.0 adds `waitForRun`, `verifyWebhook`, `subscribeFormatRun` and `uploadFile` - `/pricing/api` rate card and Korean docs ## v0.2.61 — August 2, 2026 Teamspace, Format handles, Agents polish - Personal and Team agent lists are separate - Invite email leads straight into the team - Faster switch between Personal and Team - Custom Team and Member settings - Formats use the `{handle}/{slug}` identity; Format cards use 4:5 - Billing errors (402) show apart from upstream errors ## v0.2.60 — August 2, 2026 Formats, docs, org settings - Starter Formats install automatically - Format cards use 16:9 with a Featured row - B-roll prefers reference-to-video; a single Seedance reference image routes there too - Team and Member settings pages - Docs add Formats and Completions pages and a Changelog page ## v0.2.59 — August 1, 2026 Next 16, Agent Completions, Scheduled, team wallet - Public **Agent Completions** API: `POST /v1/agent/completions` with run receipt endpoints - Product chrome retitled **Actions → Scheduled** (API path stays `/v1/actions`) - Org / team **pooled wallet** and team concurrency controls - Agents polish: Home sidebar scope, Assets Images tiles, Scheduled Clock icon - System theme across product surfaces. The docs Appearance default is System - Web stack unified on Next.js 16 ## v0.2.58 — August 1, 2026 Auth route policy, vanity Formats, onboarding, SDK scaffold - Stronger auth route policy with clearer sign-in / sign-up destinations - Real sign-up on `/sign-up`. Auth redirects keep the locale - **Vanity Format** invoke URLs (`handle` + `vanity_invoke_url`) - Guided creation / onboarding flow - Private Hey API TypeScript client scaffold - Agents ⌘↵ hydration fix and Pretendard unicode-range scoping ## v0.2.57 — July 31, 2026 Formats API, vanity Actions, team chrome, Dashboard ko-KR - Formats API improvements and vanity Action invoke URLs - Team workspace chrome wired to real org collaboration - Account / team handles in Settings and Agents - Dashboard UI localized for ko-KR ## v0.2.56 — July 31, 2026 Action media harvest, Skills chrome/authoring, ko-KR Agents - Action API harvests run output media from the job ledger - Skills gallery chrome and in-chat authoring polish - Agents product chrome localized for ko-KR - Dashboard polish and reliability fixes ## v0.2.55 — July 31, 2026 Action API 503, Grok 4.5 default, custom output schemas - Action API upstream 503 fix (no longer misreports generation runtime) - Grok 4.5 as the default agent model - Custom output schemas for structured agent results - Actions docs coverage and Agents toaster polish ## v0.2.54 — July 30, 2026 Action API trigger, live-commerce S5 lock, MCP TTS/i2v - Action API-call trigger settable and runnable in Agents UI - Live-commerce S5 visual lock - MCP tools for TTS and image-to-video - Skills gallery scroll and accent CTA contrast fixes ## v0.2.53 — July 30, 2026 Agents Action/Skills chrome, Action API runs, timeline - Agents Action and Skills chrome alignment - Action API runs surface in product - Timeline media improvements - Dark default theme and navigation loading polish --- Source: GitHub Releases tags `v*` (only product releases, without `skill-lab-v*` and `open-v*`). To refresh this page, use `gh release list` / `gh release view vX.Y.Z`.