How it works

The mental model is simple: KiCad is home, Pinflow is the AI co-designer that sits next to it. Pinflow doesn’t replace KiCad — you keep your project open in KiCad and use Pinflow to add and edit circuitry quickly.

The two-pane shell

Left — the chat agent. You describe what you want (“add a 5V→3V3 buck,” “resolve real parts for this block”). The agent can read your currently-open KiCad schematic, draft and place subcircuits, and look up orderable parts.
Right — the live schematic viewer. It renders your open .kicad_sch and updates as the agent proposes changes, so you see what’s happening as it happens.

Review before it touches your files

This is the part worth internalizing: the agent’s edits don’t go straight to disk. Changes land in a staged working copy that you review in the viewer, then explicitly commit or discard.

While staged, the viewer highlights what changed.
Commit writes the change back to your real .kicad_sch.
Discard throws the working copy away — your file is untouched.

Nothing is written to your files without an explicit accept. Pinflow never writes into your project’s git history — a commit is a plain file write to your .kicad_sch.

Bring your own key

Pinflow runs on your own Anthropic API key (see Getting started). The agent’s reasoning happens through that key; your schematic and prompts go to the Anthropic API to power the agent, and nowhere else unless you opt into Pinflow Cloud.

Pinflow is in alpha. The agent loop, the live viewer, and the staging workflow are real, but many individual agent capabilities are still being built. When a feature isn’t implemented yet, the agent is designed to say so rather than pretend.