What's New

Claude Opus 4.6 — Anthropic's latest flagship model is now available across all plans and in the free sandbox.

o3 (full) added — OpenAI's o3 reasoning model joins the lineup alongside GPT-4.1 and o3 Mini.

Model picker refreshed — removed outdated entries (Claude 3.5 Haiku, GPT-4.1 Nano) and reordered all providers so the newest models appear first.

Grok 3 label updated across provider tables and blog content to match xAI's current branding.

Free Sandbox — test a live AI assistant at /try without creating an account. Bring your own API key (Anthropic, OpenAI, Google, xAI, or DeepSeek), pick a personality, and start chatting in 30 seconds.

BYOK (Bring Your Own Key) support — paste any supported API key and it's validated in real-time. Keys are never stored; they exist only in server memory for the duration of your session.

Personality templates — choose a pre-built assistant persona (developer helper, writing coach, data analyst, and more) so the sandbox conversation is immediately useful.

Sandbox sessions are capped at 50 messages and 30 minutes to keep shared resources fair. A live countdown and message counter keep you informed.

Real-time capacity indicator shows how many sandbox slots are available before you start, so you know instantly whether the service is ready.

Navigation updated — "Try Free" link added to the site header on desktop and mobile, highlighted in the accent color for easy discovery.

Browser Automation promoted to Beta — the browser tools are now production-ready with improved reliability, per-action timeouts, and automatic retry on transient failures.

Plugin permissions — blog and Reddit tools are now gated by plan with clear upgrade prompts, so free users aren't left guessing why a tool didn't work.

All browser actions now enforce per-action timeouts (navigation 60 s, screenshot 30 s, snapshot 10 s) — no more requests hanging indefinitely if the browser service is slow.

Reddit and blog publishing now automatically retry on transient network errors (502, 503, timeouts) with exponential backoff, reducing false failures.

Large tool outputs (browser snapshots, screenshots) are now intelligently truncated to keep agent context lean and response times fast.

Plugin manifests updated to v1.1 with tool annotations — agents now know which tools are read-only vs. write operations.

Improved browser session reliability — login flows on Reddit, X, LinkedIn, and other platforms are now significantly more stable with better network routing and session handling.

Two new platforms added to Browser Automation — Product Hunt and Hacker News are now available in the platform picker.

Browser sessions now use improved network routing and environment configuration for more reliable page loading and interaction.

Platform list across the app is now always in sync — no more errors when selecting a platform in the dashboard that the backend doesn't recognize.

Fixed an issue where selecting certain platforms (like X or Product Hunt) in the browser session picker would fail with a validation error.

Fixed browser snapshot fallback for pages that don't render the interactive element tree correctly.

Browser Automation (Beta) — your AI bot can now open browser sessions and interact with platforms that don't have APIs. Log in through a secure viewport in the dashboard, and your bot handles the rest. Supports Reddit, LinkedIn, X, Product Hunt, and more.

Reddit Integration — your bot can now post and reply on Reddit. Configure subreddit targets and let your agent engage with communities, answer questions, and share content.

Google Workspace Channels — connect your bot to Gmail and Google Calendar. Send emails, read your inbox, create calendar events, and manage your schedule through your AI agent.

Hashnode added as a supported blog platform — your bot can now publish blog posts to Medium, Dev.to, and Hashnode directly from chat.

New reusable modal component across the dashboard — cleaner UI, consistent animations, and better keyboard accessibility for all dialogs.

Dashboard now shows a What's New section and announcement banner so you can discover new features without leaving the app.

Total Blocking Time reduced from 4,140 ms to near-zero — the Three.js WebGL hero scene is now loaded via requestIdleCallback (4 s timeout) so the main thread stays free during initial page interactivity.

Hero parallax scroll effect now mutates the DOM directly instead of going through React state, eliminating the main source of long tasks during scroll.

Gzip compression enabled for the self-hosted Next.js server — HTML, JS, and CSS payloads are now 60–80% smaller on first load.

Hero section and all below-fold content are now server-rendered HTML — client-side hydration is limited to interactive islands (3D scene, typewriter, parallax), cutting JavaScript execution on load.

CTA pulse animation is now GPU-composited (opacity only). Previously it animated box-shadow, causing a repaint on every frame. No visual change.

lucide-react icons now use package-import optimisation — only the icons actually used are bundled, reducing JS weight.

Fixed Content Security Policy (CSP) blocking the Cloudflare Web Analytics beacon (static.cloudflareinsights.com) and Microsoft Clarity script loader (scripts.clarity.ms) and pixel (c.clarity.ms).

Added <main> landmark element to all public pages (landing, login, register, channels, status, 404, error) for WCAG 2.4.1 compliance and improved screen-reader navigation.

Published new blog post: Why Your OpenClaw Bill Is So High (And How to Cut It by 90%) — a practical guide covering context limits, compaction, prompt caching, and model routing with real cost numbers.

Interactive OpenClaw cost calculator — estimate your monthly API spend across four optimisation scenarios and generate a ready-to-paste config snippet. Available at /tools/openclaw-cost-calculator.

Cookie consent banner — visitors can now accept or decline analytics and advertising cookies. Declining means no tracking cookies are set.

Homepage load time reduced — the 3D hero animation is now loaded after the page content is visible instead of before, so text and CTAs paint immediately.

Static assets now served with long-lived cache headers so repeat visits load significantly faster.

Status page now shows a full 30-day uptime history for each service, colour-coded day by day, plus a summary of current health across all services.

p95 latency bars — the status page now shows real response-time performance for each service with a progress bar and target threshold so you can see not just if things are up, but how fast they are.

Incident detail view — past incidents now include a root cause summary, affected services, and a chronological timeline of updates.

New comparison page: Clawship vs Self-hosting OpenClaw — a side-by-side breakdown across deploy time, channels, security, monitoring, and cost. Find it at /vs/managed-openclaw-hosting or via "Compare" in the navigation.

Main navigation updated — "Compare" link added for visitors evaluating Clawship against running OpenClaw themselves.

Footer reorganized into labeled columns (Product, Compare, Support, Legal) for easier discovery.

Blog publishing is now a paid feature — available on Starter plans and above. Free plan users see a clear upgrade prompt in the dashboard.

Blog Publishing added to the pricing page and landing page features grid so new users can discover it before signing up.

Pre-commit hook now auto-fixes formatting and import order issues using Biome before every commit, preventing lint errors from reaching production.

Fixed blog publishing failing in certain deployment configurations — resolved a networking issue that prevented the publish callback from reaching the backend.

Fixed a plugin configuration warning that could appear in logs when the blog publishing tool was loaded.

Updated the underlying bot engine to the latest security-patched release.

Fixed WebSocket connection issues that could prevent the dashboard from communicating with deployed bots in certain network configurations.

WebChat — chat with your AI bot directly from the dashboard. No developer setup required — just open your instance, click "Open Chat", and start talking.

Fixed WebChat message delivery — conversations now stream responses correctly in real time.

Fixed a configuration issue that could prevent instances from starting in certain setups.

Fixed a rare server crash on startup — logging now falls back gracefully in production environments.

Fixed an intermittent date parsing error that could cause unexpected failures.

Improved production deployment reliability — dependencies are now handled more cleanly at runtime.

Connection errors now return friendly messages instead of raw server errors.

Blog Publishing — connect your DEV.to or Medium account, and your AI assistant can write and publish articles directly from chat. Supports draft and publish modes, markdown formatting, and per-platform tag limits.

Blog management dashboard — connect/disconnect blog platforms, view connected accounts, and manage API keys from your instance settings.

Accurate instance status — instances now show a "Starting" state while booting up, instead of jumping straight to "Running" before the bot is actually ready.

Instance health detection — the dashboard now verifies your bot is actually running when you view your instance. If it crashed or was removed, the status updates immediately instead of staying stuck.

Readiness checks — a background process monitors each deploy and only marks the instance as running once the bot is confirmed responsive.

Fixed an issue where instances could appear stuck in a loading state indefinitely if the underlying process was removed outside the dashboard.

Fixed blog API key validation — keys are now verified against the platform API before being saved, so invalid keys are caught at connection time.

Published new blog post: 12 real-life use cases for AI Telegram bots — covering productivity, education, business, and creative use cases.

Blog system now supports multiple posts with full SEO metadata, JSON-LD structured data, and dynamically generated Open Graph images per post.

Free plan — deploy a Telegram bot with any AI model at $0/month, forever. Gated by Discord community membership.

Shared instance architecture — free plan users share resources efficiently while maintaining full isolation of their bot conversations and configuration.

Discord OAuth integration — seamless community verification flow. Join the server, verify your account, and your free plan activates automatically.

Initial platform launch with deploy wizard, instance management dashboard, and Stripe billing across four plan tiers.

Support for 10+ messaging channels: Telegram, Discord, WhatsApp, Slack, Signal, iMessage, Microsoft Teams, Matrix, WebChat, and Google Chat.

Multi-model support — deploy with Claude, GPT, Gemini, or bring your own API key for any supported provider.

Encrypted token storage — all API keys and bot tokens stored with AES-256-GCM encryption at rest.

10 programmatic SEO channel pages with unique content, structured data, and Open Graph images.