Skip to content
Cognolo

How Cognolo works

Cognolo reads your software, asks the owner what the code cannot say, and proposes the actions an assistant may take as a pull request. Nothing can be called until a developer merges it. Each step below says how far it has got.

The loop, in seven steps

The approval step runs today, inside one real product. Most of the others are built and tested, but none has been used for a customer yet, and nothing is hosted. See it move on the Home page.

  1. 1

    Understand

    Cognolo lifts off your app and reads the code: who may do what, what each action changes, and what can be undone. It never opens a secrets file, and nothing it reads is run.

    Built, and run on one real app

  2. 2

    Check the docs

    What it learned goes into a context file: facts from the code, each with where it found them, and notes of its own about what it spots. Checking it against your docs comes next; where they disagree, the code will win.

    Partly built

  3. 3

    Ask the owner

    From what it noticed, it asks you plain questions. A clear answer is saved as a rule in your own words. If it is unclear, Cognolo asks again.

    Built, and tested with the real model

  4. 4

    Propose tools

    It draws a map of your app, lists the actions an assistant could take, and sends them back as a pull request. A plain check reads the code again with the change in place before anything is sent, and the owner can switch any rule off.

    Built, not yet used for a customer

  5. 5

    Act

    A developer approves the pull request. The assistant now works inside your app and asks before it changes anything.

    Runs today in one real app

  6. 6

    Get feedback

    When someone tells the assistant what is missing, it thanks them and writes a one-line request, which they see. If they click Send, it reaches the owner, who puts it in their own words and can turn it into a proposed change. Their own words are not sent.

    Built, tried with a test cloud

  7. 7

    Stay in sync

    A developer changes the code. Cognolo checks the pull request against what it knows and flags any skill the change could break.

    Built, tested against a stand-in for GitHub

The cloud plans. Your app does the work.

The part inside your app runs today, in one real app. The cloud's parts are built and tested, but not hosted yet.

Diagram of three areas. Your repository on GitHub holds your code and the tool pull request. The Cognolo cloud holds the code reader, the knowledge map, and the signed rules and activity log. Your app holds the assistant panel, one API route, the merged tools and your database, and its route talks to the AI model.Your repositoryon GitHubCognolo cloudreads and keeps the rulesYour appdoes the workYour codeTool pull requestCode readerKnowledge mapSigned rules and activity logThe AI modelyour key or oursAssistant panelOne API routeruns tools as the signed-in personMerged toolsYour database
  1. 1. Connect

    A GitHub App lets Cognolo read your code and build a cited map of it. It only reads, unless you let it open pull requests. Files that hold secrets, such as .env files, keys and certificates, are never read.

  2. 2. Propose

    Cognolo proposes the tool layer as a pull request. Nothing can be called until a developer merges and deploys it.

  3. 3. Act

    A person asks in your app. Your app asks the AI model, with its own key, checks every action against the rules Cognolo signed, and runs the merged tools as that person, against your own database. Credentials and database access never leave your app.

  4. 4. Record and keep up

    Every lookup, proposal, approval and decline is recorded, with no customer data unless you choose to include it. When your code changes, Cognolo checks each pull request against what it knows and flags what it would break.

Who does what

Engineers

  • Install the Cognolo GitHub App so Cognolo can read the repository. It opens pull requests only if the owner allows it.
  • Review the pull request that proposes the tools, and merge it when they are happy.
  • Install the package from npm, run npx cognolo init to add its files, add the keys and deploy. An AI coding assistant can do these steps for them.

Owners

  • Answer the plain questions Cognolo asks where the code is unclear. A clear answer becomes a stored rule; a vague one gets a follow-up question.
  • Read what each skill did, and decide how much the assistant may do alone, separately in Dev, QA and Production.
  • Watch the assistant's activity log, kept apart from what people did by hand.

Three ways to start

The plan is to add trust one level at a time. None of these is available as a hosted product yet.

  1. Level 0

    Connect

    Read-only. Cognolo builds the map, checks the docs and asks its questions. Nothing runs in your product.

  2. Level 1

    Staff assistant

    You merge the generated pull request, add a key and deploy. Your staff get the assistant.

  3. Level 2

    Customer assistant

    You open roles and tools to your own customers and set how much the assistant may do alone.

Want to shape it with us? Early access is by conversation.