Changelog

What's New

All the latest updates, improvements, and fixes to rtrvr.ai. We ship fast—join our Discord to stay in the loop.

Join Discord for Live Updates

v35May 15, 2026

Real-Time Steering for Rover

Guide an in-flight Rover task without restarting it — typing or voice.

New rover.sendFeedback(text, opts?) API — send mid-run guidance to the agent without starting a new task. Each item goes Queued → Applied (with step index) or Dropped (run ended, canceled, or queue full). Returns { id } if accepted, null if no run is active.See example
Composer stays enabled mid-run in steer mode (typing and voice). Placeholder switches to 'Guide Rover (this won't start a new task)…' so the distinction is unmistakable, and mid-run dictation no longer auto-stops.
Feedback is sanitized and bounded before injection — control chars stripped, 16 items per run cap, 500 chars each, triple-backticks escaped — so visitor guidance can't inject prompt directives.
Voice narration now reliably speaks the agent's ask_user questions and final responses when the visitor used the mic — fixed a string-equality check that silently dropped the voice signal on the auto-submit path, plus a follow-up trace that wasn't re-expanding for second messages in the same chat.
Tool-result narrations (e.g. 'Email sent successfully') were silently filtered at the scheduler entry and never spoken even with voice on; they now route through the same TTS queue as tool-start narrations.

v34May 8, 2026

Better Tool Building, More MCPs, and Shareable Rover Demos

Private Mode, refreshed credits, generated tools, recordings, Rover onboarding, and tighter extension-cloud handoff across the whole stack.

1-click connect MCP connections — HubSpot, Notion, Slack, Github, Zapier, Google Calendar.
Private Mode is now wired end-to-end — Disable logging and accessibility-tree storage, keep storage extension-side, and upload files as bytes for private runs and non-Gemini providers.
Credits now refresh monthly instead of incrementally stacking, with pricing copy updated to match the new model.
Email-link sign-in is now the default auth path, with backend protections that block password-based signups.
Generated tools are easier to inspect and reuse — Shared task views, detailed task views, schedules, and triggers can now surface and run tools created from recordings or cloud workflows.
Captured recordings now show up across cloud panels and task details, with indexed task executions and a private task-viewing bridge from the extension.
Custom tools can now do more on their own — They can create sheets, append columns and rows, read webpages, reuse connected tools, call MCP tools, and turn page content into clean structured results.Tool docs
Tool generation is faster and more resilient, with temporary generated tools, improved script generation, and cleaner schemas for complex MCP servers.
Subroutines are easier to reuse — Save useful actions as shortcuts, pin generated tools, trigger them from MCP clients like Claude or ChatGPT, schedule or trigger them from the cloud, and keep improving them over time instead of rebuilding from scratch.
Smarter page targeting — Generated tools now use more reliable page signals like roles, labels, and accessibility data, so they are less brittle when a site changes.
Extension reliability fixes — Stopped duplicate text insertion on LinkedIn and Instagram, fixed Calendar MCP ping failures, reduced frequent shortcut searches, and captured documents even when recording had just started.
Sandbox runs are easier to trust — Sandboxed act/extract now has a centralized 2.5 minute timeout, agentic tab groups are shown during execution, and temporary execution tabs are cleaned up afterward.
The extension onboarding path got smoother with a welcome carousel, multi-step shortcuts, pinned tools that also appear as subroutines, and minted Firebase token handoff between the extension and website.
Rover can help AI assistants act on your site — ChatGPT and other assistants can discover Rover instructions, hand off the task, and let Rover complete the action in a cloud browser.Read the post
Shareable Rover demos — Send a URL that opens your site with a Rover task preloaded, so customers, teammates, or followers can try an automated flow directly.
Rover onboarding and runtime configuration were refreshed, with faster site configuration, narration fallback, and cleaner install/runtime behavior for embedded agents.

v33April 15, 2026

AI Subroutines, BYO ChatGPT/Claude Subscription & WhatsApp Control

Zero-token deterministic automation, plus a wave of Rover updates

AI Subroutines — Recordings now intercept network requests as you go. Save them as deterministic tools the agent calls with zero token cost and 100% repeatability. Built for IG DMs, X posts, LinkedIn invites, and the long tail of site-specific actions.
Reuse your ChatGPT or Claude subscription — Sign in with OpenAI or Anthropic OAuth and drive rtrvr with the plan you already pay for. No second API bill, no key wrangling. BYO API key still works.
Control rtrvr from WhatsApp — Trigger /schedule, /trigger, and /check_schedule_results commands directly from a chat thread. Same shortcuts work from the Planner and MCP.
Knowledge Base cloud scrape + index — Add URLs from the dashboard and we'll scrape and index them straight into a Knowledge Base.
New MCP tools — List stores, query, and index Knowledge Bases directly from ChatGPT, Claude, or Cursor. Plus list recordings, schedules, and Subroutines.
Accessibility Tree clipping — MacCtrl+C copies the full Agent Accessibility Tree of any tab. Paste into your KB or any LLM for richer context than plain text grabbers.
Voice input — Dictate prompts and recordings via on-device Chrome transcription.
Microsoft Office File Support: DOC/DOCX, .XLSX, .PPT file formats supported.
Auto-share to Sheets — Scrape results stream straight to a shared Google Sheet, no extra clicks.
Simplified UI — One + button for files, tabs, sheets, and Knowledge Bases. Action toolbars collapsed to a hidden utility bar. Welcome screen suggests templates based on your current URL.
Every cloud panel ships an API snippet — Re-run any execution programmatically with one copy-paste.
Faster onboarding — Min instances on sign-in/out functions to cut latency on the first-touch path.
Agent analytics — PostHog-style visibility for AI agents. Built at the Zero to Agent hackathon (Vercel × Google DeepMind). Track visits, runs, trajectories, an AX Score for your site, agent feedback, and memory notes so agents learn instead of rediscovering friction.Read the post
Agent Web Protocol Stack — Our research thesis on how agents actually access the web and why it needs an execution layer, not just a reading layer.Read the thesis
Protocol vs. Prompt Injection — Breakdown of the Mintlify incident and why declared protocols beat clipboard injection.Read the analysis
Vibe code your own agentic Chrome extension — Fork our Rover preview-helper and ship a browser agent without building the agentic stack from scratch.GitHub
Coming next: a public library of community Subroutines, templates, and datasets. Plus Private Mode ($99/mo+) for zero data residency — in-memory only, nothing logged.

v30March 19, 2026

Vibe Hacking, Open-Source Rover, and Gemini Flash Lite by Default

API-native scraping, agentic websites, and cheaper free daily driving

Vibe Hacking — Agents can now reverse-engineer the real APIs behind websites, discover GraphQL/XHR calls from page interaction, and generate reusable scripts for scale scraping.Read the launch post
Our early sandboxed code-execution bet now powers network-level scraping workflows, including replaying requests with your browser auth context for lower-friction extraction.
Roadmap: reusable network scripts as tools, cloud execution for repeatable scale, and hybrid workflows that combine network calls with DOM parsing.
Rover is now open source — turn any web interface agentic and integrate it into your own browser automation stack, extension, or Electron app.View on GitHub
Coming next for Rover: WebMCP exposure, site-specific action/API indexing, recorded demos, voice-powered accessibility flows, and agent-consumable site actions via URL parameters.
Gemini Flash Lite is now the default for lightweight tasks, and Google AI Studio's 500 requests/day should cover most free daily-driving use cases.

v29February 26, 2025

Product Hunt Launch, Bring Any LLM Provider, & Easier MCP Setup

Launch day updates and extension flexibility upgrades

Rover is live on Product Hunt. Upvote support unlocks a 1-month free promo code for Product Hunt usersView Product Hunt launch
Bring ANY LLM provider in the Chrome Extension by adding your own API key (OpenAI, Claude, OpenRouter, Moonshot, MiniMax, DeepSeek, Z.ai, and more), with automatic fallback to Gemini on failures
Local model support in the Chrome Extension — point to any OpenAI-compatible chat completion endpoint (like Ollama, LM Studio, or vLLM) for zero API-cost runs
Direct MCP Server setup in Extension Tools dropdown — connect providers like HubSpot, Notion, Slack, and GitHub without config files or terminal steps
For MCP servers that do not support dynamic registration, you can now enter client ID and client secret directly in the extension

v28February 19, 2025

Rover Launch, AI Watchdog Triggers, Knowledge Bases & 90% Cost Reduction

Biggest release yet — meet Rover, the world's first embeddable web agent

Rover — World's first embeddable web agent. Drop a single script tag on any site and give it a conversational AI that clicks, fills, navigates, and converts — right inside your UITry Rover
Rover use cases: onboarding copilots, support agents that reproduce issues in-app, growth flows that turn "I'm stuck" into "done." Have a custom use case? Reach out for a partnershipRead the deep dive
90% cheaper agent runs — Skip the planner and directly choose sub-agent modes: act, extract, or crawl. Plus Gemini Flash Lite 3 support, better token caching, and heuristic pattern matching for data extraction
Browser Triggers — AI Watchdog sits in your browser and fires workflows on tab notifications. Get an X/LinkedIn DM → trigger a workflow to respond. Perfect for sites that don't expose webhooks but do push browser notificationsDocs
Knowledge Bases — Turn open tabs into searchable AI memory. Add tabs, ask questions, refresh sources anytime, and share via an embeddable chat widget. Use your credits or deduct from yoursDocs
Knowledge Base live example: AI India Summit — see a shared knowledge base in actionSee live example
Cookie Syncing — Sync cookies between Chrome Extension and Cloud Platform. Select which domains to sync in Settings — cloud browser agents stay logged in seamlessly
rtrvr-cli — Trigger rtrvr from the command line. Works with OpenClaw, includes rtrvr Skills for enhanced performanceGitHub
MCP has more powers — Everything you can do in the Extension is now exposed via MCP/CLI: list tools, recordings, knowledge bases, and create tools programmaticallyCLI Docs
WhatsApp Bot — Trigger your browser automations straight from WhatsAppTry it
Webhooks — Fire a webhook after any workflow completes. Chain workflows, push to Slack, update your CRM — whatever you needDocs
Refreshed docs at rtrvr.ai/docs, better markdown rendering in chat responses, and lots of quality-of-life improvements
Rover Product Hunt launch coming 2/25 — subscribe to get notified and leave a review!Get notified on Product Hunt

v26January 16, 2025

BYO Multiple API Keys, Gemini 3, & Workflow Templates

Major upgrade with Gemini 3 models and one-click workflow templates

BYO Multiple API Keys — Add multiple API keys that auto-rotate on rate limits. Pro tip: Bring Gemini keys from 5+ Google accounts = basically free daily web agent usage
Gemini 3 Upgrade — Now running on Gemini 3 Flash and Pro models
Workflow Templates — One-click workflow templates, no more prompt engineering hassleBrowse templates
Cloud VNC Display — Watch your parallel cloud browser agents execute in real-time
Extension as API/MCP: Replay Workflows (deterministic execution, share & rerun via URL), File Support (pass URLs or upload files), Recording ID (reuse recordings)
Chat UX Updates — New hints for fresh chats, popup suggestions, Cmd + Shift + Enter = instant new chat
OAuth MCP Support — Just use mcp.rtrvr.ai, no API key needed. Call cloud /scrape and /agent APIs directly from MCP
File Pills — Visual file pills for better context control
Incremental Google Auth — Request drive.file access only when needed
Agent Performance — Smarter Sheets handling, improved orchestration, better sub-agent accuracy
Native captcha iframes, shadow DOMs handling, Comet sign-in fixes, new onboarding flow
New blog posts publishedRead blog

January 6, 2025

Live VNC, Native Captcha Solving, & Parallel Execution

Live VNC into cloud browsers — Watch exactly what agents do. Take control mid-task.
Native captcha solving — No external services. We read the challenge structure directly.
Parallel execution at scale — 53 browsers tested, can go to 100+
rtrvr.ai/retrieve — One-click templates. Describe what you want, get structured websets back.Try it
Shareable results — One link. Your extracted data, ready to send.
Gemini 3 Pro and 3 Flash are live on Cloud with fallback to 2.5

v25December 20, 2024

Scheduled Monitoring & Enhanced Page Parser

NEW Scheduled Monitoring — Continuously scrape a page and append rows to a sheet on schedule. Can configure to reuse existing tabs.
More comprehensive data extraction — Updated page parser to handle ShadowDOM, cross-origin iframes
Remote triggering via API/MCP is now much more reliable
Planning updates — Much better at longer tasks and orchestrating Sheets Workflows
Max Parallel Tabs = 1 will now reuse existing tabs for Crawl/Sheets Workflows
Fixed subscription page issue on Extension
Created Sheets no longer auto-added to tab selection
Fixed Adaptive Mode for Shortcuts
Fixed referral code redemption — Get FREE 250 credits per referral!
Fixed Gemini Invalid Argument issue when falling back from your API key to rtrvr credits
Agent can solve basic captchas now!

v24December 3, 2024

File Upload Tool & Extension API Trigger

New upload file tool — Agent can reuse any file uploaded in the chat onto a webpage! End-to-end job applications now possible.
Trigger Extension via API — Directly trigger agentic actions via curl commandView docs
Hard launching API — Execute API takes a prompt and performs multi-step actions
Revamped tree construction — Less resource usage, more coverage
Added Gemini key validation, fixed crawl sub-agent, UI fixes, revamped shortcut UX

November 26, 2024

API Endpoint Hard Launch & N8N Integration

Hard launching the API Endpoint!API docs
Cloud API from N8N integration demoWatch demo
Remote MCP triggering from N8N demoWatch demo
Example workflow: Calendar event → rtrvr.ai MCP → LinkedIn reconnaissance → contextual outreach → cross-platform actions

v22November 21, 2024

Tools Re-enabled, Credits Revamp, & Agent Actuation

Tools are re-enabled — Connect MCP servers, create tools or have agent create tools on the fly and reuse
Credits fixes — Revamped calculation and rendering for accurate display
Sheets workflows now write rows as soon as being processed
Scheduling tasks is fixed
Agent actuation dramatically improved — No more double actions (e.g., double commenting on IG)
Better agent chat response messages on errors, warnings, cancelling
Chat input container is now resizable — Drag up to expand, double-click to reset
Gemini API keys now validated instantly with visual feedback
Clearer settings dropdown with referral program explainer
Fixed billing portal for existing subscribers

v21November 13, 2024

FREE MCP Tools & Settings Refresh

NEW FREE MCP tools: getPageData, takePageAction — Drive agentic browser actions with your own chatbot subscription, no credit use from us!
Granular control over which MCP Tools to expose
Settings Panel UI refresh and Gemini API Key saving fix
Add images/files via drag and drop into chat or file picker
Fixed authentication on Non-Chrome browsers — No longer asks for sign-in every time
Directly rerun an executed workflow via Play button on model messages
New shortcut modal for creating/editing/sharing shortcuts

November 12, 2024

🚀 Official Product Hunt Relaunch — Cloud/API/MCP

Breaking the browser free with cloud platform launch

Your Browser Becomes an MCP Endpoint — Copy/paste your rtrvr.ai MCP URL into any agent. True agent interoperability!
Reuse Your AI Subscriptions — Connect rtrvr.ai to Claude/ChatGPT via MCP. Command your chatbot to drive browser actions for FREE.
Trigger Workflows from ANYWHERE — API calls, web chat, webhooks (Zapier/Make), WhatsApp
Scale to Thousands of Parallel Cloud Agents — Spin up hundreds of cloud browser agents instantly
Record Locally, Re-execute in Cloud — Use Chrome Extension to record a task, then replay it reliably in the cloud

Want real-time updates? We post all changes to Discord first.

Join Our Community

New Feature

Improvement

Bug Fix

Announcement