Brainfile is a maintained operating system for Claude Code — the rules it works by, the memory of how you work, and the checks that stop it going quietly wrong. It is built for owners who want to run their business on AI and think about infrastructure never.
Every capability on this page carries its real delivery state, including the ones not wired into your pull yet — 1 of 6 are there today. We would rather you read that here than find it out later.
This is not a mockup. It is a real excerpt of the sanitized operating file this business runs on, served from this site.
### Law #19: Build-and-Verify (EP directive 2026-03-20) After building ANY script, feature, or pipeline component, verify the FULL chain — not just "it exists": 1. **Runs:** Does it execute without errors? (`python3 script.py` succeeds) 2. **Output correct:** Does it produce the expected output? (Spot-check, not just "no errors") 3. **Consumed:** Is the output read by something downstream? (Dashboard, email, another script) 4. **Wired:** Is it scheduled to run automatically? (LaunchAgent, cron, pipeline hook) - **"Script exists" ≠ "script runs" ≠ "output consumed."** All four must be true. - Root cause: 3+ `ep_had_to_ask` entries from build-and-forget pattern (ITC pipeline, dashboard generators, scripts not wired to cron). - After EVERY build task: verify all 4 links in the chain before marking done.
A capable assistant with no memory of how you work will do what you literally asked. The same assistant, running on your Brainfile, does what you actually needed — because the rules it loaded remember your setup and make it check its own work before it says done.
1 · It remembers your business. It catches your mistakes.
Done. I opened the homepage, found the price, and changed it to $49.
Your prices live in one source file, and the pages are generated from it — hand-editing the HTML would be wiped on the next build, so I updated the source.
Rebuilt the 3 pages that display a price.
Then checked the rest: an FAQ answer still quoted the old $99. Fixed that too.
All four places now read $49 — confirmed on the live pages before calling it done.
2 · It stays current.
Nothing tells you. The setup keeps running the old way until something quietly stops working, and the first sign is the thing that didn't happen.
Your next pull carries the updated pattern, and the weekly freshness note says in plain English what changed and what it means for your setup.
Two illustrations of what the shipped pattern files and the weekly freshness feed produce — the same kind of rule you can read in full above, applied to a real request, and the maintenance that keeps it true afterwards. Not testimonials, not mockups.
You can download a set of rules today for nothing. Here is the honest argument for paying for one instead.
Claude Code ships roughly weekly. A configuration pointing at a retired model or a removed setting does not throw an error — it quietly stops working, and there is no way for a non-technical owner to know. Perishability is the whole reason antivirus is a subscription.
Your decisions, preferences and context build up in plain files on your own disk. After six months that folder is genuinely yours and could not be recreated in an afternoon. It is the one asset in this product a competitor cannot copy, because it is not ours.
A rule you downloaded is inert. A check that fires while you work is a service that is either on or not — and when it lapses you notice during the work, not in an email. That is the difference between content you own and a service you renew.
Everything below is live right now and verifiable. The gap between this list and the capability list underneath it is the honest state of the product.
This is the actual payload, not a mock-up:
# Brainfile update — v9 (2026-08-24)
Merged from the Brainfile Freshness Feed. These are cadence updates to apply to how
Claude works in this project.
## Apply these patterns
- START-HERE.md orientation added to every role pack: Add a START-HERE.md at the root of your brain/ project so anyone (or future-you) opening the folder cold gets oriented in 30 seconds. Template pattern: install -> confirm-it-loaded prompt -> first personalizations -> how it compounds.
- C-suite role packs (CMO/CFO/CSO/CTO/COO) added, fed by a shared engine: Structure your own brain/ by SEAT/ROLE, not just by project: a short mission, 3-4 focus areas, and a small set of copy-paste prompts per role. Then feed all of them from ONE source file so improving a prompt improves every place it appears.
## Awareness (Claude Code / tooling changes)
- Bug fixes and reliability improvements (https://code.claude.com/docs/en/changelog)
- Cost estimates (`/cost`, status line, `--max-budget-usd`) now include the 1.1× US-only-inference premium for data-residency workspaces (https://code.claude.com/docs/en/changelog)
- Added the one-time fullscreen renderer offer on Bedrock, Vertex, Foundry and other previously excluded setups; new installs there now start in fullscreen (https://code.claude.com/docs/en/changelog)
… (full feed ships in the pack)
This is the real v9 feed — the latest one shipping in the pack (2026-08-24). The next feed ships with the next real change — we never backfill fake updates.
This list is generated from our internal feature registry. A capability only appears here after it has been proven on our own operation, verified as a customer deliverable, decoupled from anything private of ours, and cleared for honesty. Each one carries its true delivery state.
1 of 6 of these reach your machine today. The rest run in our own production and are being wired into the pull. We would rather publish that number than let you find out later.
Your computer stays awake when your assistant is working and sleeps when it isn't — set once, on a schedule you choose, without touching a single system setting.
Always-on checks that stop made-up facts, invented quotes, silent failures and slow drift — so when your assistant says it's done, it is.
Every week your system learns what changed — new capabilities, retired settings, things that would have quietly broken — and updates itself.
Before anything goes live it's scanned for leaked files, personal data, placeholder text, dead links and broken checkout — and blocked if it isn't ready.
A single live dashboard built from your own numbers, where every figure carries a threshold, an owner and what to do when it turns red.
An exportable record of every decision your system gated — what ran, when, and why anything was blocked — ready to hand to an auditor.
These are part of the plan and are not in the product. Nothing on this page charges you for them.
Starter is the running system. Pro is the running system tuned to you and to your bill. The split is generic versus personalized — never feature count.
Announced price — 2 months free at $490/yr. Starter opens when the launcher ships.
No card, no charge — an email when it opens.
or $999/yr — $83/mo billed annually.
Card only at the end of the trial. Cancel any time and keep every file.
Not ready for a subscription? We audit your existing Claude Code or agent setup and send back a scored report and a prioritized fix list, worst first.
Most trials end with a locked door. That is a bad fit for a product whose value is a folder of your own decisions — so ours is designed the other way round.
30 days, no card, email only. When it ends you keep every file you pulled, permanently. There is nothing to uninstall and nothing to hand back.
A 14-day reverse trial. At day 14 nothing is revoked: you keep every file and your entire accumulated memory. What changes is that the feed pins, the weekly report stops, and the checks go read-only — they still detect, and instead of acting they tell you what they would have blocked, and that renewing turns them back on.
Planned, not built. It ships with the launcher and the weekly report.
The hardest part of shipping a system like this is that it runs on hardware we will never see. So we are building an opt-in program: when a check fails on your machine, the shape of that failure can help harden the same check for everyone.
You will be able to see exactly what would be sent before anything is, and turn it off at any time without losing a feature.
One email when Starter opens, when the launcher lands, and when the fleet program opens for opt-in. Nothing else.
Double opt-in. Unsubscribe in one click. We never sell or share your address.
You keep them. All of them, permanently — the rules, your accumulated memory folder, and the last freshness feed you pulled. It is plain text on your own disk and we will not try to take it back. What stops is the running service: the feed, the weekly report and the active checks.
No. Brainfile runs inside your own Claude Code on your own machine. Your files never leave it. The only thing that crosses the network is the pull that fetches your setup, which is authenticated with a licence token and sends us nothing about your work.
No, and deliberately so. There are no zips and no download links on this site. Your setup is pulled from a licence-gated endpoint each time, which is what lets it stay current without you ever downloading, installing or merging anything.
No. That is the entire design constraint. Anything about your computer that can be detected is detected rather than asked, and you should never be asked for a number you would have to go and look up. If you can use Claude Code, you can use this.
The one-word launcher, the weekly report, automatic machine configuration, the two-tier split and the fleet telemetry program. They are named on this page and marked as not available. What is live today is the trial, the cloud pull, the pattern files, the role packs and the weekly freshness feed.
You should not, if a static file is what you want — there are good free ones and we point at ours below. You pay for the part a file structurally cannot do: staying current with a platform that ships weekly, accumulating your business in a way that compounds, and running checks that are either on or not.
One click in the billing portal, or email [email protected] and we will do it. No retention call, no exit survey, and the refund policy is published.