>_ Changelog

Komos now speaks MCP. Connect Claude Code, Claude Desktop, Codex, Gemini CLI, or any MCP-compatible agent to run automations directly from your editor.

5 tools available:

• list_tasks — Browse your automation tasks
• run_task — Execute a task with inputs
• get_run — Check run status and results
• list_runs — View recent task runs
• ask_moss — Talk to Moss, your AI automation engineer. Create tasks, configure integrations, set up schedules, and more.

Works with Claude Code, Claude Desktop, Codex, and Gemini CLI. Setup guide | For AI

• Longer conversations are now more reliable with improved context management and memory
• Moss better retains earlier context when working through multi-step tasks across many turns
• Faster response times for follow-up messages in active sessions

• Moss can now run long-running tasks in the background while you continue working
• Background jobs handle multi-step agent work like research, data processing, and complex automations
• Moss notifies you when a background job completes and delivers the results in chat

• Moss reads your current screen to give context-specific answers about tasks and runs
• Take screenshots of your workspace to help Moss debug automations visually
• Prompt diff review shows side-by-side changes before you apply AI suggestions in the builder

• Describe what you want to automate in plain language and Moss builds a complete task for you
• Upload a video walkthrough or paste a process document and Moss generates the automation from it
• Ask questions, iterate on task designs, or troubleshoot a failed run through natural conversation
• Runs in a secure cloud sandbox with full access to your workspace, sessions persist between conversations

• Skills are reusable instruction sets that teach automations how to handle specific workflows
• Encode team knowledge so tasks follow your processes consistently every time
• Share skills across your organization so every team member gets the same automation quality

• Komos now automatically selects the best AI model for each request based on task complexity and context
• Simple actions route to fast, lightweight models while complex reasoning uses more capable ones
• AI usage tracking and projected billing visible alongside action steps in the dashboard

• Connect tasks to 100+ third-party services: Slack, Salesforce, Jira, webhooks, and more
• Trigger integrations directly from the task builder without writing code

• Agent Actions use Chrome DevTools Protocol for more precise clicking, typing, and data extraction on complex pages
• Every task step is checked by an AI verifier that confirms the expected result before moving on
• Network request capture, JSON processing, and code execution for richer data extraction

• Run automations on your own Windows or macOS machines to reach intranet portals, legacy desktop apps, and anything behind your corporate network
• Runners authenticate via your existing SSO provider and appear as an execution target alongside the cloud sandbox

• Tasks have an assigned owner, controlling who can edit or delete shared automations
• App loads significantly faster with static asset serving across fourteen pages

• Scheduled tasks support active hours and days-of-week filters to skip triggers outside business hours
• Email notifications for task runs with separate toggles for creators and requesters

• Describe browser tasks in plain language instead of scripting every click
• Document parsing extracts structured data from PDFs, images, and scanned documents with OCR support

• Record any browser workflow or upload a video walkthrough, and our AI turns it into a reusable automation
• Visual task builder with managed cloud sandboxes and real-time run monitoring
• AI-powered selector healing keeps your automations working when websites change their layout