What your team works out should stay with your team.

One person's win becomes everyone's head start. Rema is evidence-grounded memory for engineering teams. It reads the tools your team already writes in, answers with the exact passage, and keeps what someone corrects so nobody solves the same problem twice.

Five reasons

  1. 1

    The model is the same for everyone.

    Your competitors rent the same models from the same vendors. What they cannot rent is what your team already decided, and why. That record is the advantage, and today it is scattered across threads, tickets and pull requests nothing reads together.

  2. 2

    Every answer shows where it came from.

    An answer is the exact passage with the file and the lines it came from, so anyone can open the source and check it. When a newer thread overrides an older answer, the newer one comes back with it.

  3. 3

    A correction is kept, not thrown away.

    When someone fixes an answer, the fix is stored with the evidence it came from, so the next person starts from it. The idea was tried first on a bug-fixing agent, where every closed bug became a case the next one could find.

  4. 4

    It says what it did not read.

    A system that reads part of a file and reports on all of it is wrong in a way that looks careful, and reviewers rarely catch it. In Rema, a claim about every item has to rest on complete coverage, name its sample, or say how much it actually saw.

    The bug that taught this
  5. 5

    It stays yours.

    The quick start runs on your machine, on a public file, with no API key. What your team knows does not leave to make someone else's model smarter.

    Read the quick start

Two fair questions

Is this not just retrieval over our documents?
Retrieval is the easy half. The part that compounds is the record of what a person decided when the evidence was ambiguous, kept next to the evidence, which no index gives you on its own.
What if our documentation is bad?
It usually is. The decisions are rarely in the documentation; they are in pull requests, threads and tickets. Those are the sources.

Where it is today

The install is one command. The hosted console is in a closed pilot with a few early teams. It reads from Slack, GitHub, Jira, Confluence, Notion, Figma.