Piebald is here!

Launch!

After months of development, we've finally made our initial release of Piebald, v0.1.0!

Piebald is a fully functional agentic development tool—open folders, chat, get stuff built. You'll be relying entirely on Claude's, et al., ability to handle agentic tasks; the system prompt by default is empty, and it's left for you to craft completely on your own. (Access it via the chat sidebar, a gear icon in the top-right corner.)

If you prefer battle-tested setups, Piebald will soon enable you to emulate different tools like Claude Code and Gemini CLI, using their prompts and builtin tools. But for those who want to completely take control over how the agent behaves, Piebald is there for you to express yourself to the utmost; you won't be held to existing sets of instructions and principles (except for any potential server-side injected prompting by your provider of choice).

You can configure hyperparameters like temperature, top P, and stop tokens, as well as more abstract configuration like reasoning effort and prompt caching. As providers have evolved from model inference hosts to AI platforms, low-level knobs like temperature are less relevant; GPT-5, for example, requires the temperature to be 1.0 exactly. But if you're experimenting with local models or training your own LLMs, being able to configure all the nitty-gritty is a must.

You can also create "profiles"—reusable collections of configuration like system prompt, hyperparameters, and higher-level configuration. They're like "agents" or "presets" in other apps you might be familiar with.

Next steps

MCP "support" like you've never experienced is in the works. For the brief present, only builtin tools for file reading/writing/editing, file name and file content searching, and running terminal commands are available to the model. Web searching and fetching will be out in a few days.

Piebald's got some rough edges, and being so new, it's missing some features you'd expect from Claude Code, Gemini CLI, and the rest. We're aiming to reach feature parity with the leaders in this space within the next few weeks, so if you're disappointed that we don't have sub-agents, or plan mode, check back soon!

We'll harden the frontend implementation with thorough integration tests—the backend is already covered—to keep (unpleasant) surprises to a minimum.

After that, it's all new features, and that is in your hands. Our goal is to make your life as an agentic developer easier, and hearing your pain points with existing tooling and your vision for a uniquely tailored workhorse is the best way to achieve that goal.

So, what features do you want? What would make or break Piebald for you? Check your open Claude Code issues, Gemini CLI issues, Codex issues, if you had something but you can't remember it just now. Post your writeup(s) to our issue tracker on GitHub. We're all ears!

Sincerely yours,
the Piebald team