How it works
The loop that drives your app.
Every request, job, and exception your app sends goes around the same loop. Most of it runs without you today. Each stage below says what ships now and what comes next.
The loop, stage by stage
6 stages-
01
Observe
TodayThe rails_mind gem reports requests, background jobs, exceptions, business events, and customer accounts, each tagged with its release. Ingest validates and redacts before anything is stored.
-
02
Find
TodayExceptions are grouped. A request that repeats one query many times, takes over a second, or got slower after a release becomes a finding, with its evidence attached.
-
03
Diagnose
TodayAn investigation reads the evidence and your source at the release that served it, through the GitHub App, and writes down a suspected cause. Once you turn automation on for an application, new and returning exceptions start one on their own, inside a daily budget you set; you or your agent start one for any other finding.
-
04
Brief
TodayA fix brief gathers the evidence, the source, the analysis, your repository's test conventions, and a definition of done into one document for a coding agent.
-
05
Change
TodayClaude Code, or any agent, follows the brief: a failing test first, the narrowest fix, the full suite, then a pull request. You review it and merge.
ComingA runner for every workspace, so RailsMind opens the draft pull request itself.
-
06
Verify
TodayA finding with no new evidence for a week (the default) resolves quietly, and reopens if the problem returns. A daily digest lists new findings, regressions, and finished investigations.
ComingA measured before-and-after for the release that shipped the fix, with a warning when there is not enough comparable traffic to say.
LaterFixes roll out behind a Flipper flag and roll back on their own when the numbers get worse.
Autonomy
Five levels of self-driving.
Borrowed from the levels for self-driving cars. RailsMind is at level 3: it notices, investigates, and writes the brief; your agent writes the change; you merge it.
-
Level 1
It shows you
Today
Dashboards and a daily digest. You notice the problem, find the cause, and write the fix.
-
Level 2
It tells you
Today
Findings with evidence: grouped errors, suspected N+1s, slow requests, and releases that got slower.
-
Level 3
It hands your agent the fix
Today You are here
It investigates against your source and writes a fix brief. Your agent writes the change; you review and merge it.
-
Level 4
It opens the pull request
Coming
RailsMind drafts the change itself, then measures whether the release fixed the problem.
-
Level 5
It ships and steers
Later
Fixes roll out behind a flag and roll back on their own when the numbers get worse.
Guardrails
Built to be trusted with the wheel.
-
A person merges
RailsMind never merges or deploys. A brief ends at a pull request that a person reviews.
-
Leads, not proof
A finding is evidence and a hypothesis. Every brief asks for a failing test before any code changes.
-
A spend circuit breaker
Model calls stop for the day at a platform-wide cap, checked before anything is requested, and each application's automatic runs stay inside the budget you set.
-
Redaction at ingest
Payloads are validated and redacted before they are stored, and the gem can keep exception messages inside your app.
-
Access you choose
The GitHub App sees only the repositories you install it on, and each application reads one. API tokens belong to one workspace and can be revoked.
-
No code runs by default
RailsMind runs your code only in an isolated runner, and only when the operator configures one. Fix briefs need none.
Get started
Take your app for a drive.
Create a workspace, add an application, and connect it with one key. Sign in if you already have one.