Runchat Updates: May 4, 2026

Schedule autonomous agents, optimize model costs, and build complex apps with our new coding sandbox.

RunchatMay 4, 2026
release-notes

⏰ AGENT TASKS

  • You can now define agents that run on their own on a schedule on https://runchat.com/dashboard/tasks
  • Each agent has an initial prompt, chat config and workflow to use as a scratch pad
  • Agents can run workflows and check their work, leave notes for you or themselves, work together in teams and a bunch more
  • We're already using agent tasks for <@1482615099972915251> and will be sharing these with the community for anyone who wants to build their own marketing agents.

📺 TUTORIAL SERIES

  • We now have a full Runchat 101 series on youtube and the docs
  • This makes it super easy to learn runchat from the ground up

🌸 DEEPSEEK PRO AND COST/SPEED TOGGLE

  • You can now set a toggle to prefer the cheapest or fastest provider for models hosted on openrouter
  • This is especially useful for models like deepseek where the price difference is 1/5th the price of the most expensive provider
  • Deepseek is extremely cheap. You can generate workflows or write python scripts for ~10 credits.

⌛ MORE POWERFUL CODING SANDBOX

  • The improved sandbox allows agents to write much more powerful and flexible code using any node / javascript api
  • It also paves the way for creating full custom apps backed by databases, using external libraries and rendering multi-page web apps to the canvas (soon...)

🏷️ LABELS AND OTHER UX POLISH

  • Nodes now have labels. This makes it easier to tell which nodes the agent is referring to when you're creating workflows with the agent
  • Workflow privacy is now displayed on the dashboard
  • We now remember your chat settings so they persist across chats
  • Artifact nodes show the published url be default
  • Agents can read images from urls
  • Double click zooms to nodes without opening the image editor
  • Team members can switch to their own personal subscriptions when team credits expire

Whats next? More powerful agents, apps and sandboxes...