Pliant workbench is not a workbench. It is the infrastructure to build your own and keep changing it.
Every app ships a fixed idea of how you should work. A browser to read. An editor to write. A terminal to run. Slack and Zoom to talk. A chat box for AI. Optimizing the workflow between them is hard.
Pliant flips this. Your workbench is never finished. You describe the workflow you want, your AI builds it on Pliant, and you reshape it tomorrow.
"When I select text on a page, open a scratch pad next to it. Let my agent watch it and help while I type."
That's not a feature we ship. It's a workbench you build in one sentence.
| Browse | A real Chromium engine. Web apps like Slack and Zoom included. |
| Edit and run | An editor and a terminal, shared with your agent. |
| Your agent, at home | Bring the agent you already have, with its memory. It uses Pliant through MCP. |
| A UI you define | A native AppKit DSL. Every panel, trigger and flow is yours. |
| Safe to change | Every change previews first. Apply, reject or roll back. Nothing breaks for good. |
You and your AI decide what the workbench is. Pliant handles the hard parts: data, processes and stability.
Early. A macOS demo is working: a Pliant-owned Chromium embedder and a browser whose layout users redefine with preview → Apply / Reject → restore.
Next: a browse + edit loop. Send text from a page to a note, then jump back to the source.
Editor, terminal, agents and the full DSL are designed, not built. → Design · Decisions · Architecture · Build
macOS first · Apache-2.0 · by Yifan Li · Share an idea