> ## Documentation Index > Fetch the complete documentation index at: https://docs.cto.new/llms.txt > Use this file to discover all available pages before exploring further. # Changelog # September 3rd 2026 ### Sites and domains * More complete site publishing experience, with clearer separation between preview and published sites. * You can now buy domains directly through cto, connect your own domain, and manage domains from your business settings. * Added custom domains for agent email inboxes. ### Chat and agent teams * Added suggested next messages. * Live tool calls and tasks now appear directly in chat, making it easier to follow what your team is working on without switching views. * Improved activity indicators so it is much clearer when an agent is actively working. * Improved approvals, chat recovery and automatic resumption. ### Agent capabilities * Added web search support across more models. * Improved tool discovery and made connected integration status clearer for MCPs and GitHub. * Made Secrets easier to find and improved the wider business settings experience. ### Models * Added recommended models throughout the product, including during business creation, model selection and after upgrading. * Added newer models including DeepSeek V4 Pro 0813, Fable 5.1, Muse Spark 1.3 and GLM 5.3. * Unified chat and image model handling and improved model capability, pricing and intelligence information in the picker. ### Ads and plans * Plus and higher plans are now ad-free. ### Mobile and UX * Continued improvement of the mobile experience. * Improved the site empty state, lead chat input, task readability, markdown rendering and settings navigation. * Added clearer email warm-up limits and other inbox status information. ### Reliability * Improved automatic recovery for broken or stranded sandboxes and reduced cases where a working environment could incorrectly appear unrecoverable. * Added self-healing for chats affected by interrupted or orphaned tool calls. # August 5th 2026 ### Product and UX * Added “Surprise me” option to start a new project. * Improved marketplace browsing with price filters, clearer listing layouts, sales visibility, and improved mobile support. * Improved loading states, task views, approval cards, the organisation chart, and navigation across smaller screens. ### Agent teams * Rolled out a new agent-team architecture with improved token efficiency, smarter delegation, and more stable working environments. * Improved approvals so users can provide rejection reasons, continue chatting while approvals are pending, and automatically clear stale approvals. * Added stronger task recovery and more reliable handling when agents are paused, fired, restarted, or interrupted. ### Sites and custom domains * Added stable public sites, publishing controls, and support for custom domains. * Improved the site tab with more space, deep links, shared editing permissions, better caching, and clearer recovery when a preview stops responding. * Improved handling of ports and running processes so users should need to think less about how their site is hosted. ### Integrations and project ads * Added sponsored ChatGPT ad campaigns that can be created and managed from inside cto. * Improved GitHub and MCP access for delegated agents, along with clearer integration and permission controls. ### Billing, usage, and models * Added the option to purchase a one-off usage reset after reaching a limit. * Improved the model picker with clearer costs, recommendations, and model change handling. * Added and updated models including Claude Sonnet 5, GPT-5.6 Luna, DeepSeek V4 Flash and Pro, and newer image-capable models. * Retired a number of older models and now move affected chats to suitable replacements more reliably. ### Reliability * Improved sandbox recovery, including automatic retries, mid-task replacement, volume recovery. * Improved long-chat handling through earlier compaction, better stream buffering, and limits on excessive terminal output. * Fixed cases where chats could freeze, messages could disappear, previews could remain blank, or tasks could spin indefinitely. * Improved GitHub authentication, custom-domain reliability, model switching, inbox safeguards, and error messages throughout the product. * Added more resilient background task processing so delegated work can continue without blocking the main chat experience. # June 28th 2026 ### Product and UX * Added stable public websites for agent teams, with support for publishable `ctonew.app` URLs and custom subdomains. * Added default websites out of the box. * Improved the business view with clearer navigation, better integrations/MCP surfaces, recent tasks, finance widgets, mobile polish, and more prominent GitHub/repo actions. * Added marketplace listing files, editable listings, seller finance views, listing sales counters, and clearer paid/free marketplace listing behavior. * Added clearer usage-limit modals in the business view, including warnings before limits are reached. * Added better model switching from the org chart. ### Chat, agents, and websites * Improved long-running chats with worker threads, better compaction, better terminal-output caps. * Added chat reset and recovery improvements for cases where agents get stuck with stale context. * Improved lead-agent behavior around plans, approvals, finance, and email. * Added stronger guardrails against hallucinated business information. * Added support for agents to create Stripe products and payment links. * Added better website handling around ports and previews. * Added more reliable file downloads and “download all files” flows. ### Models * Added and updated models including Gemini 3.5 Flash, MiniMax M3, Kimi K2.7, DeepSeek V4, GLM 5.2. * Removed or deprecated older models including GLM 4.7/5, MiniMax M2.5, Kimi K2.5, Devstral. ### Finance and email * Added Stripe Connect improvements for finance views, payment links, and AI businesses balances. * Improved email handling with per-team scoping, abuse guardrails, and automated cooldowns. ### Integrations and MCPs * Added MCP search, connected-MCP listing, and better MCP permissions/setup handling. * Added GitHub and repo connection improvements, including clearer repo creation and sync actions. ### Reliability * Added sandbox backup, restore, full migration, manual upgrade, recovery buttons, and recovery. * Improved sandbox health checks, transient error detection, restart behavior, host masking, workload containment, and file recovery. * Improved chat/API reliability with worker isolation, memory fixes, terminal-output limits, websocket backpressure handling, and better error surfacing. * Improved Turso retry handling, logging, and resilience. # June 5th 2026 ### Product and UX * Launched the first version of the CTO CLI. * Added private marketplace listings, editable listings, AI-generated listing descriptions, clearer seller-country support, and better listing markdown. * Replaced business “goals” with clearer business plans. * Added a lead chat model switcher, reset chat, stop button, better task viewing, improved repo access, and clearer handling for pending approvals in Agent Teams. * Expanded mobile support across AI businesses, recent tasks, reset chat, business views, and marketplace flows. * Added recent tasks to the main business/headquarters views. ### Chat, agents, and models * Improved chat compaction, pagination, scrollback, and message storage so long-running business and agent team sessions are less likely to lose context. * Added DeepSeek V4 Flash, Claude Opus 4.8, Gemini 3.5 Flash. * Fixed agent image generation, Gemini/Anthropic image handling, and large file/image memory issues. * Improved Agentmail inbox provisioning, webhook handling, safety controls, and paused-team behavior. ### Integrations and developer experience * Moved MCPs/integrations into business views, and improved setup copying, permissions, and compatibility with the latest MCP SDK. * Improved repo access handling, connect-repo prompts, and marketplace/business repo flows. ### Reliability * Added sandbox backup, recovery, upgrade, and migration flows. * Improved sandbox health checks, restart behavior, workload containment, process handling, and observability. * Improved handling for provider timeouts, client disconnects, memory pressure, and websocket backpressure. * Reduced memory usage across chats, images, previews, and agent sessions; moved more message history out of large JSON blobs. # April 30th 2026 ### Product and UX * Create and run teams of agents with the AI business project type. * Browse, buy, and sell agent teams on the marketplace. * Agents can now receive email and wake up when new messages arrive. * Shareable achievements for agent and app milestones. ### Chat, apps, and previews * Better preview recovery, restart handling, proxy behavior, mobile previews, and file downloads from agent team file browsers. * Improved chat scrollback and storage so longer conversations and team activity remain visible and load more reliably. * New models including GPT-5.5, Claude Opus 4.7, Kimi K2.6, GLM 5.1, MiniMax M2.7, and Grok 4.20. ### Integrations and tooling * Better MCP assignment, permissions, and setup handling. * Improved Convex dashboard and Expo app generation/preview support. ### Reliability * Improved sandbox cleanup, persistence, restart behavior, and abuse prevention. * Improved handling for provider rate limits, model-specific errors, and image-heavy chats. # April 4th 2026 ### Product and UX * CTOClaw early release: build and run general purpose agents with no code. * Build mobile apps with Expo. * Agents can now receive email, with wake-on-email support. * Web app publishing and sharing. * Move repos between projects, with clearer project type selection and better defaults. * Improved restart/recovery for web app previews and deployments. * Improved mobile app and small-screen experiences across the product. ### Chat and apps * Added slash commands in chat. * Chat and manual tasks now support image attachments. * Agents can now use a browser during runs. * Expanded model selection and model switching across web apps, mobile apps, and agent flows. * Improved chat compaction so longer sessions continue more reliably. ### Reliability * Improved preview reliability, websocket stability, and recovery from temporary failures. * Better handling for common failure modes. # March 3rd 2026 ### Web app projects * Added web apps as a project type. Includes a live preview and interactive chat for vibe coding. * Provides Convex for backend services. * Sensible default development environment and settings for ease of use. * Automatically deploys and configures TanStack codebase. ### Product and UX * **Projects & workspaces:** Organize repos and apps into projects, with improved onboarding and nicer defaults. * **Chat quality:** Better long-chat handling via compaction so sessions can continue without hitting context limits. * **MCPs:** Manage MCP connections from the web UI, with clearer re-auth prompts when needed. ### Billing & limits * Increased monthly usage limits across all paid tiers. ### Models * Added/updated models including Claude Opus 4.6, Gemini 3.1 Pro, GLM-5. * Allow model selection for free users. # February 6th 2026 ## Product and UX * Start building without linking a git provider. * Better handling of common failure modes. * Chats are now “agent sessions” with clearer scoping between sessions. * Cleaner task logs with improved tool-call rendering and live status. * Tasks started from chat now stay scoped to the current session. * Prevent duplicate messages while a chat run is already in progress. * Automatic team naming. ## Infrastructure * Migrated sandboxes to Daytona for faster startup and responsiveness. ## Integrations * Added Supabase and Vercel MCP presets. * Improved MCP auth/setup reliability. ## Models * Added GPT‑5.2 Codex. * Added Kimi K2.5. # December 23rd 2025 ☃️ ## Live Previews (beta) * Live preview available in beta to all users. * Start new template project from cto.new. * Add build and run commands to task runner. * Setup agent attempts build and run commands. ## Smarter model selection * Route tasks based on quality, speed, capacity, and cost. * Expanded model support, including: * GPT-5.2 * Sonnet 4.5 - GPT 5.2 alloy * Claude Opus 4.5 * Gemini 3 Flash preview) * GLM-4.7 * MiniMax M2.1 * Devstral 2 ## Engine-agent chat improvements * Faster, more reliable task planning. * Better visibility into tool usage during runs. * Tasks started from chat are now properly scoped and isolated. * Simplified planning and draft task UX. ## MCP overhaul * Unified MCP system with clearer setup and fewer edge cases. * Global MCPs can now be configured once and reused in agents and chats. * Per-chat MCP selection for more control. * Improved support for HTTP, SSE, and stdio MCPs. * Better error handling and clearer tool naming. ## Reliability & performance * Faster VM startup times. * More robust VM lifecycle handling. * Better handling of temporary provider outages. ## Benchmark * Introduced cto bench, measuring real-world task success rates across models. * Live chart and leaderboard updated every day. ## UI & UX polish * Mobile improvements. * Keyboard shortcut: ⌘ + Enter throughout the app. * Create empty repo from cto.new. * Clearer task and preview states. * General responsiveness and stability improvements. # November 26th 2025 ## Product & UX * Add "get code" buttons for easily copyable git checkout commands for a task. * Add task preview in alpha. * Clearer draft PR/MR status. * Chats moved to their own page. * Add privacy mode for zero data use. ## Models & Routing * Added GPT-5.1 and GPT-5.1 Codex and removed GPT-5 and GPT-5 Codex. * Added GPT-5.1 Codex Mini. * Added Gemini 3 Pro Preview. * Added Kimi K2 Thinking. * Added MiniMax M2. * Improved and updated auto model routing. ## Infra, Events & Git * Harden events queueing and retries. * Update GitHub webhooks and improve git reliability. * Status page: [status.cto.new](http://status.cto.new). # November 15th 2025 ## Models * Added auto-model selection. * Haiku set as chat default for improved speed and tool calling. * Tasks started from drafts use preferred model. ## Chat * Optimistic send for chat messages. * Cancel chat messages. * 20k character message limit. ## UI/UX * Git checkout command in task view. * Fix chat text area resizing, mobile fixes, dark-mode tweaks. * Surfaced secrets error messages. ## Auth and Invites * Update invite eligibility to prevent abuse. * Google sign-in restored. ## Reliability and Bugfixes * Fix diff saving. * Race fix on user/org creation. * Validate env/secret format. * Fix “ghost queued” tasks. # October 31st 2025 🎃 ## Chat & UX * Fixed chat text area resize. * Optimistic send - show user messages instantly. * Chat summaries: agent can now reference last 10 chats. ## Models & Memory * Added memory to chat agent. * Improved chat label generation. * Added Haiku, improved Anthropic prompt caching. # October 29th 2025 ## Tasks & Workflow * Shorter draft titles. * Chat agent now aware of tasks. ## Models & Execution * Anthropic fully restored (incl. Alloy). * Add GLM-4.6. * Improve prompt caching. ## Abuse & Rate Limits * 20k-char cap on first chat message. * Domain blocks for obvious botting. * Added CAPTCHA and bot blocks. ## Git & Repos * Hardened git token fetching/retries. ## Reliability & Ops * Removed legacy ticket plans/speculative mode; removed context7 MCP and Anthropic web\_search tool. * Start routing tasks to Northflank for scalability. ## UI * Dark mode. * Text area resize fixes. # October 22nd 2025 ## Access & onboarding * Invite UX: add invite button; generate all available codes. * Team invites: invited team members no longer need codes. * Sign-up protections to block bots. ## Draft tasks & task UX * Prevent multiple tasks per draft. * Save draft edits on start; show actionable controls only when valid. ## Models & routing * Add AWS Bedrock for Anthropic model failover. * Better labels & long-chat handling; include model name in system prompt. ## GitHub & repos * Fix Start GitHub issue from @cto tag. * Branch selector no longer crashes if branch deleted. ## Reliability & safety * Fewer flaky runs: increase events timeout; harden Git token fetching/retries; empty-repo/default-branch fixes. * Auto-ban abusive agent chat calls (unknown adapter/invalid schema).