CODE INTELLIGENCE FOR SHIPPING TEAMS
Know what your product actually does.
Ask how a feature works today and you get four answers — the ticket, the documentation, somebody’s memory, and the code. Only one of them is true. Releap reads that one, keeps up with it, and answers from it.
Works with GitHub and Azure DevOps · Jira, Azure Boards, GitHub Issues, Aha! and Notion connect as sources, and every write-back is opt-in and per integration
The problem
Nobody can tell you how it works now.
Not because your team is careless. Because the answer is spread across a ticket written before the work started, a document last edited in March, a Slack thread nobody can find, and eleven thousand lines of code.
Every team hits this at the same four moments, and every team solves it the same way — by finding the one person who remembers, or by reading the source.
A new engineer joins
Two weeks of reading code because the documentation describes a version that no longer exists.
A PM writes a spec
It contradicts current behavior, and nobody notices until the ticket is in review.
Support answers a customer
“Is this a bug or intended?” — and the honest answer is that nobody knows.
You come back to old code
Six months later, and the reasoning is gone even though you wrote it.
Why it keeps getting worse
The faster you ship, the less anyone writes down.
Six years ago a change took two weeks and three people read the diff. Now it takes an afternoon, and the ticket that described it was written before anyone knew what the code would have to do.
The gap is not bugs — tests catch bugs. The gap is decisions: the ones made in the editor because they had to be made, and then never recorded anywhere a person could find them. Somebody asks why exports cap at 50,000 rows and nobody remembers, because nobody decided it in a meeting. It was decided in export/filter.go, on a Tuesday.
Not in the ticket
The ticket describes the intent, not the twelve things the implementation had to settle.
Not in the docs
Documentation describes last quarter, and nothing tells you which page went wrong when the code moved.
Not in anyone’s head
The person who decided it moved on — or it was an agent, and that reasoning was in a context window that no longer exists.
What Releap does
Answers questions from the code. Catches what the code decided.
Two halves of the same job. Ask it anything about how the product works today — and when something changes, it notices before you have to.
Ask — and get the current answer
“Why do exports cap at 50,000 rows?”
There is a row cap in export/filter.go:88, added because Postgres times out above it. Marco confirmed that was intended on 2 August. REL-4877 makes it configurable, shipping 12 September.
Four sources, one answer, every one of them current.
Catch — when the code decides something new
A merge lands. Releap compares what shipped against what the ticket said, and asks about the difference.
Three decisions this change made that REL-4884 does not describe.
One click and it becomes part of the answer above, permanently.
REL-4884 — Suppression rules merged. 3 decisions the ticket does not describe:
• Suppression is scoped to the workspace that created it suppress/rules.go:88
• Blocked entirely on security-classified decisions verify/classify.go:142
• A written reason is required, no format check suppress/reason.go:34
Not saying any of these are wrong. Only that the ticket does not describe them.
[ All fine ] [ Not what I wanted ] [ Needs thought ] [ Not a decision ]
One click and it becomes a confirmation — a permanent record that a human said this behavior is intended, attached to the code that implements it. That record is the thing everything else is built on.
How it stays true
Every answer makes the next one better.
Most documentation tools decay because writing them is a separate job nobody has time for. Releap’s record is a by-product of shipping: somebody confirms a decision once, in a Slack message, and it is available everywhere afterward.
Ask
Ask why exports cap at 50,000 and the answer cites the code and Marco’s confirmation from 2 August. Not a guess from a wiki page.
Know
A confirmation that contradicts a published page flags that page, names the section, and drafts the correction from the confirmation.
Comms
Release notes written from what was confirmed, not from commit messages. Unconfirmed changes are held back rather than described wrongly.
Listen & Support
Somebody asks whether behavior is intended. There is a recorded answer with a name and a date on it.
This is why the modules are one product rather than a bundle. Verification produces the record; everything else reads it. A documentation tool with no verification is guessing, and a release-notes tool with no verification is paraphrasing commit messages.
On agents
When an agent implements it, the reasoning is already gone.
A person can be asked why they did something. An agent’s reasoning lived in a context window that has been discarded. The code is the only surviving record of what it decided.
In our own repository, agent-authored changes carry about 2.3× as many decisions the ticket does not describe. That is our number, from our code — not an industry figure. Not because the work is worse: an agent settles ambiguity instead of asking, and settles it silently. Releap measures the ratio for your repository and reports it.
This is not a tool for reviewing AI output. It reads every diff the same way regardless of who wrote it. Agents just make the problem arrive faster.
Point it at a repository and see.
It replays your last 90 days before it sends anybody a single notification.