Skip to content
Agent CityAgent City

What's new

The changes you’ll notice in each feature release, newest first. Agent City updates itself, so you’ll get these as they come out. Bug-fix releases in between are listed in the full release notes on GitHub.

0.28

Section titled “0.28”

October 2026

  • Telegram, Signal and X are installed from Browse. They’re no longer part of the app. Install them from Agents → Browse, then connect them on Connections. See Messaging agents.
  • Scout is built in. The web agent is now part of Agent City: always there, and you can turn it off but not remove it. It never sees your files. See Scout.
  • Web search without a key. Scout and the System agent search through DuckDuckGo when there’s no BRAVE_API_KEY. See System.
  • Lists under the reply. When an agent answers My team with a list or table, it’s drawn under the reply with its links. See Chats.
  • Feedback. The Feedback button in the top bar reports a bug or suggests a feature (0.27.2). See Troubleshooting.
  • For agent authors. An agent can sign in its own way with SIGNIN, announce urgent messages with ALERTS, and name the oldest Agent City it needs with requires:. See Agent files.

0.27

Section titled “0.27”

October 2026

  • Long-running agents. Ask for research (“do some market research on…”, “research … in depth”) and the agent works on it in the background while you carry on chatting. A card shows its progress, Stop has it write up what it found, and the report is posted to the chat and saved in your workspace. See Long-running agents.
  • Steps per agent. Each agent’s settings have a Steps setting, from 5 to 200: how many times it may use its tools for one job.
  • Everything at once. Chats, automations, events and Reflection time now run side by side instead of queuing behind each other. Cloud models answer up to 4 requests at once (Settings → Models → Working at once). See Choose a model.
  • Post to Activity. Ask any agent to “post to Activity when you’re done”, and it does. See Activity.
  • Experimental tab. Reflection time has moved to its own Settings → Experimental tab (0.27.1). See Reflection time.

0.26

Section titled “0.26”

October 2026

  • Edit an added model. Under Settings → Models → More models, Edit changes a model’s provider, server, name or key in place. Agents using it pick up the change straight away. See More models.
  • A tidier Agents page. It now opens on a compact list, one row per agent with its model and status. Cards are a click away. See Agents.
  • Utilities moved. Shared memory, Connections, the Audit log and Settings are now under the spanner at the bottom of the sidebar.
  • Pin and search chats (0.26.1). Pin a chat to keep it at the top, and use the search button beside Chats to find one by name. See Chats.

0.25

Section titled “0.25”

October 2026

  • Keys stay with their agents. Each agent’s tools can only see the keys and sign-ins it declares. An agent installed from a source asks for your OK (Allow on Connections) before using a key you saved for another. See Connections and keys.
  • Reflection time (experimental). At the end of the day, the agents that worked meet in a little voxel office to look back on it and each save one lesson. See Reflection time.
  • Free talk fills the silence. While the agent works on what you said, it says something short like “On it”, so the wait isn’t silent. See Voice.
  • Quick-reply buttons. When a reply ends with a short question (“PDF or Word?”), the answers show as buttons above the message box.
  • Approvals don’t hold things up. While an agent waits for your OK, your other chats, phones and automations carry on.