An agent can produce a change quickly. Reviewing it still takes attention: reading the code, understanding the trade-offs, and leaving feedback where it matters.
I built Suikou to make that review a conversation. It gives me a place to read closely, attach comments to the work, and hand a review round back to an agent.
An earlier version of the review interface; the current workflow no longer uses per-file verdicts.
Review the work in context
Suikou supports both whole-file and Git diff reviews. Code gets syntax highlighting; Markdown, diagrams, images, and HTML can be viewed in a form suited to the artifact. For HTML, you can switch between reviewing the rendered page and interacting with it.
Comments can be anchored to lines of a file or to elements in an HTML preview. Each comment can say whether it needs a fix, an answer, or is simply a note. That makes the next step clearer for both the reviewer and the agent.
Work in rounds
The workflow is simple:
- An agent creates a review for files or a Git diff.
- I read the work, leave comments, and submit a round.
- The agent responds to the comments and makes any needed changes.
- I review the result and continue the discussion if needed.
Suikou's CLI lets an agent create a review, wait for feedback, and reply in the relevant threads. The human submits each round; the conversation stays alongside the work being reviewed.
Local by design
Suikou runs on your own machine as a single-user workbench. Its interface is available in a browser, so you can also reach it from another device on your network.
The name Suikou (推敲) refers to carefully weighing the words you choose. That feels fitting for a tool built around taking a closer look before a change moves on.
Suikou is open source. Explore the project and its setup instructions on GitHub.