braid

Code comprehension engine

The diff is dead,
long live the diff.

The diff never died. It just stopped being enough when agents write the code. Braid captures what you asked for, the decisions your agent made, and the checks it ran, and ties each one back to the diff.

Install Braid

You keep working in your coding agent. Braid ambiently tracks everything. When the change is ready, ask your agent to reconstruct the intent and review the result.

Why Braid?

Nobody reads code anymore!

It is not that people don't want to read code. Agents write more of it than anyone has the bandwidth to follow, and code you didn't write is harder to understand. We still want to steer the agents and own the direction and the outcome of their work, and that takes understanding what they did and why.

Braid is the missing layer. It gives you that understanding by being an ambient monitor of your agent. As the agent works, Braid records what was done, decided, noticed, and checked. It then reconstructs the intent behind your work and connects the agent's choices and checks to the code. You can follow the work, question a decision, and steer the next change without losing the high-level system context.

As a result, Braid helps you:

  1. Develop a strong understanding of what you want to merge or ship, and the evidence behind it.
  2. Make knowledge transfer in your team simpler.
  3. Rehydrate the coding loop when you come back to a change later.
Without Braid, understanding of a change drops at each handover: the reviewer starts over from the diff, and the organisation starts over from git blame. With Braid, the traces, decisions, and intent captured while the agent works stay with the change for every reader.Without Braid, understanding of a change drops at each handover: the reviewer starts over from the diff, and the organisation starts over from git blame. With Braid, the traces, decisions, and intent captured while the agent works stay with the change for every reader.

Where does Braid fit in your current workflow?

Your prompt in the coding agent session is the start. An agent may delegate parts of it to subagents, change code, run checks, and revise its approach. The final diff shows the code changes; the session also contains the context behind them, such as an issue from GitHub or a document from Notion.

Codex or Claude Code reads context such as Notion, Linear, or Google Docs during the session. On your machine, Braid captures the sessions, records the intent, and opens it in a local review UI, where you inspect and steer the agent. You can push the intent to the GitHub PR as a comment.Codex or Claude Code reads context such as Notion, Linear, or Google Docs during the session. On your machine, Braid captures the sessions, records the intent, and opens it in a local review UI, where you inspect and steer the agent. You can push the intent to the GitHub PR as a comment.

Understand what your agent has done

Reconstruct the intent after the agent has finished and committed changes. Braid sends the captured work and code changes to its server, where an AI model explains the goal, the outcome, the decisions, and the steps that led there. A braid is a record of one branch's change: the captured work and the intent reconstructed from it.

Open a decision to see the code it affected and the agent work trace behind it. Check the explanation against that evidence before you accept the change.

If you would make a different choice, tell your agent what to change and why. Have it commit, then reconstruct the same braid again. The updated intent includes your change and why. Review it before moving on.

Help your peers understand your code

Once you are satisfied with the agent's work, you can push the same intent from the coding agent session to the Braid server. Braid also posts the intent as a comment on the PR, so that your peers, or their agents, have the full context they need to understand the change you are trying to merge.

The intent stays private until you push it. Then anyone who can read the repository can open it after signing in with GitHub.

This helps build a shared understanding of any change, big or small!

Choose your next step

Early access

# you're on the list ✓# couldn't reach the server — try again