You give a clear brief
Everyone agrees what is being built.
Free for Windows and Linux
The operating system for AI software development.
Your AI coding agents plan like an architect, wait for your approval, test what they build, and share a project memory that never forgets, without re-reading the whole project every time.
>
irm https://cortexrail.com/install.ps1 | iex
Free. One line, about a minute. Other ways to get it
Try it: the demo is waiting for your approval
- Plans firstlike a software architect
- Never forgetsand never re-reads it all
- Any AI, any teammateone shared project brain
The problem · AI without Cortex Rail
Without a system, AI is fast. Then it drifts, forgets, and sends the bill.
This is what coding with a plain chat window and an AI agent looks like today. Nothing in that setup plans the product, remembers last week, or stops two people building the same thing. It is the problem Cortex Rail was built to solve.
How a project slips
The AI hits a gap and guesses
It doesn't stop to ask. It picks, and keeps typing.
A new session forgets the reasons
It reads the whole project again, and decides differently.
The guess is now the foundation
Everything built since depends on it. Undoing it costs a release.
The four things missing
No plan
Code first, design never. Nobody decided the structure; it just happened.
Cortex Rail plans first, like an architect →No memory
Every new chat starts from zero and re-learns the project at your expense.
Cortex Rail never forgets →Rising cost
The bigger the project gets, the more the AI re-reads before each small change.
Cortex Rail loads only what the task needs →Team collisions
Two people, two AIs, the same task, and neither knows about the other.
Cortex Rail gives every AI one shared project brain →
Cortex Rail fixes all four. Here is how, starting with the plan.
↓
Planning
It thinks like a software architect before it writes a line.
Cortex Rail makes the AI do what a senior engineer does first: understand the product, ask about the gaps, design the structure, decide how it will be tested, and break the work into small tasks. Only then does it ask for your approval.
- Understand
Reads what you want
Your description of the product, or the code you already have.
- Size
Fits the process to the project
A quick demo stays light. A product with real users gets the full treatment.
- Ask
Asks instead of guessing
Every gap comes back as a question, with its recommendation and the reason.
- Design
Draws the blueprint
Structure, data, error handling, security risks and how it will be tested.
- Break down
Splits it into small tasks
Each one is clear enough to build, check and hand to a teammate.
- Approve
Waits for youLOCKED UNTIL YES
The AI cannot write code before you approve. The tools refuse the write.
- Build
Builds one task at a time
Inside the structure it designed, with the board showing every move.
- Prove
Reviews, tests, then tells you
It checks its own work to the depth the project needs before calling it done.
Right-sized, never watered down
Pick the size. See what the AI prepares.
Planning
Testing
Your approvals
Security work
Always on, at every size clean code · project memory · lean context · plain words · the approval lock
Memory and context
It never forgets. And it never re-reads everything.
Most tools make you choose: a short memory, or a huge one the AI must load every time. Cortex Rail keeps every decision as a small note, understands what each note means, and hands the AI only the few that matter for the message in front of it.
Handed to the AI for this message
Loaded now
0 notes
Everything remembered
36 notes
The rest stays on disk, untouched and unpaid for, until a message needs it.
0%of the time the right note is in the top five, in our own test of 95 questions
~0 msto search, on your own computer, with nothing sent anywhere
Found by meaning
Ask about “money” and it finds the note about prices in cents. A small search model runs on your computer; no cloud search, no extra bill.
A smart index, not a big file
The AI reads a tiny always-loaded core and a map of the notes, then opens only the ones the task needs.
It writes its own notes
When a task is done, the AI saves what it learned: the decision, the reason, the pitfall. Every save has an Undo.
Survives long sessions
Before an AI squeezes a long conversation to make room, the important parts are saved and given back afterwards.
Knows when it may be wrong
If the code a note describes has changed, the note is marked “check this” so the AI looks before trusting it.
Tidies itself
Old notes are put away automatically, important ones are never deleted, and everything lives in your project as plain files.
What that does to the bill
The project grows. The cost of each task doesn't.
An AI that re-reads the project before every change gets more expensive every week. Cortex Rail loads one section of the rules, a tiny memory core and a few notes, so task number 300 costs about what task number 3 did.
- Only the rules for the current step are loaded
- Notes are fetched by meaning, a few at a time
- Updates change one line, not a whole file
Teams
Many people. Many AIs. One project that knows everything.
The memory and the task list travel with the code. So whoever opens the project, on any computer and with any AI, starts with the same decisions, the same plan and the same picture of who is doing what.
Maya Claude Code
Builds half of “Card payments” on Monday.
Sam Codex
Opens the project on Tuesday, on his own laptop.
Lina OpenCode
Joins on Thursday. Has never seen the code.
The shared project brain
memory + task list + handoffs
Plain files saved with the code. Every teammate's AI starts from them.
Card payments
0%
M
S
Maya starts “Card payments”.
No double work
Starting a task puts your name on it. Start one that's taken and you get a heads-up first: take it over, or pick another.
Handoffs without a meeting
When someone stops, a short note is saved: what they were on, what isn't saved yet, what comes next. The next person is greeted with it.
Same context on any machine
No “it works on my setup” for AI context. Pull the project and your AI knows what theirs knew.
New people useful on day one
A new teammate's AI reads the same memory as everyone else's. Onboarding is opening the project.
Many AIs
The right AI for each step. All on the same rail.
Planning is short but decides everything. Writing the code is long but only follows the plan. So put your strongest AI on the thinking, a cheaper one on the typing, and a different one again to check the result. They follow the same rules and share the same memory.
StepWhat it needsWho does it (click to change)
You choose the AI for every step. The app finds the ones you have installed and never picks for you.
Testing and security
It knows what to test before it writes the code.
How deep to test is decided with the plan, not remembered afterwards. Building, reviewing and testing are separate jobs with different questions, so the AI that wrote the code is never the only one saying it works.
Looked at from every side
The logic
Does each piece do what the plan said, including the awkward cases?
The pieces together
Do the parts still work when they are connected to real data?
Bad input
What happens when someone sends rubbish, or something malicious?
Outside services
Do we still agree with the payment, email and other services we call?
Real people
Can it be used with a keyboard and a screen reader?
The look
Did a change quietly break a screen that used to be right?
A demo gets a quick click-through. A product with real users gets all six. You choose the size; the AI does the work.
Security
Approval is a lock, not a polite request.
The AI is blocked from writing code until you approve the plan. The block is enforced by the tools themselves and again when the work is saved, so an AI can't talk its way around it.
- Risks are thought through while planning, before anything is built
- Passwords and keys never go into the code
- Bigger projects add approvals for the data design and before release
- The app runs on your computer; it has no server holding your code
The app
Run all of it from one chat, in plain words.
No commands to memorise, nothing to copy between windows. Say what you want; the app starts the right AI on the right step, shows you what it is doing, and asks when it needs you.
Everything through the chat
Start work, approve plans, check where things stand, answer the AI's questions.
A board that moves by itself
Tasks slide from Not started to Done as the AI works. Stuck ones call for you.
Autopilot when you want it
Let it write, test and check task after task, stopping whenever it needs your OK.
Yours, on your computer
No Cortex Rail account. Your AI tools keep using their own sign-ins.
Get Cortex Rail
One line. Free. Yours in a minute.
It installs into your own folders, needs no administrator rights, checks the download against its published fingerprint, and keeps itself up to date.
Paste into PowerShell:
>
irm https://cortexrail.com/install.ps1 | iex
Runs onWindows 10 and 11 · Ubuntu 20.04+, Debian 11+, Fedora 34+ and similar
Works withClaude Code, Codex, OpenCode, Gemini CLI, Qwen Code, Cursor CLI, GitHub Copilot CLI, Aider
CostsNothing. Free for personal and commercial work.