Skip to main content

Agents arrive in the Playground

This week, ToolHive Desktop UI v0.35.0 makes Agents in the Playground available to everyone, with two built-in agents and the option to author your own. The Desktop UI also retires the localhost TCP port it used to talk to ToolHive in favor of a per-user socket, and the Playground picks up message editing, queueing, and per-message cost.

Agents in the Desktop UI Playground​

Agents in the Playground are now available to every user of the Desktop UI. Each chat thread runs against an agent that sets the system prompt and a default toolset, so you can shape the Desktop UI around the task in front of you instead of starting from a blank model every time.

The Desktop UI ships with two built-in agents:

  • ToolHive Assistant for general questions and workflows across your installed MCP servers and skills.
  • Skill Engineer for going from "I want a skill that does X" to a working, tested skill without leaving the Desktop UI. It scaffolds a SKILL.md, iterates with you on the contents, and tests it against your MCP servers in the same window. When you're done, install or publish the skill from the same place.

You can also author your own agents with their own system prompts and default toolsets, and pick one per thread. Each thread keeps its own agent, model, MCP servers, and skills selection, so a Skill Engineer thread can sit next to a ToolHive Assistant thread or a custom agent thread without configuration leaking between them.

No more localhost port conflicts with the Desktop UI​

The Desktop UI no longer binds a localhost TCP port to talk to ToolHive on your machine. Instead, it uses a UNIX socket on macOS and Linux, or a named pipe on Windows. What that means for you day to day:

  • No more port conflicts when another dev tool wants the same localhost port you were using for the Desktop UI.
  • Tighter local security: the socket is scoped to your user account, so other accounts on the same machine can't reach it.

It's a behind-the-scenes change, but it removes a class of "why is my port already in use" headaches.

More Playground improvements​

A handful of smaller Playground improvements ship alongside:

  • Edit and resend a previous message. Tweak a prompt that didn't quite land and rerun from there, instead of starting a new thread.
  • See the estimated cost of every assistant message, using models.dev pricing data.
  • Queue follow-up messages while a response is still streaming, so you don't have to wait between turns.
  • One-click copy and agent-aware suggestions on the empty state, so new threads start with prompts tailored to the agent you picked.

Getting started​

For detailed release notes, check the project repositories:

You can find all ToolHive documentation on the Stacklok documentation site.