Skip to content
How it works

You describe the situation.
You get a procedure.

Not advice, not a checklist and not a prompt someone else wrote for someone else. A procedure fitted to your stack, your infrastructure and the technical decisions you already made.

Choose a plan
Browse the categories
The lifecycle

Seven steps, and you can stop after any of them

Most people run the first four and never need the rest. The last three exist for the moments when the answer actually matters.

  1. Describe what you want to do

    In your own words: “I want to know whether our deployment is safe.” The system works out which kind of technical action that is and what it will need from you.

    You can also browse the categories directly if you already know what you want.

  2. Answer only what is missing

    Your stack, hosting, integrations and previous decisions are already stored. You are asked about the gaps, not about everything again.

  3. See the cost before you spend it

    What will be produced, what it will cost in credits, and what you will be expected to bring back. Nothing is charged until you confirm.

    If generation fails, the credits are returned. A failure on our side is not your problem.

  4. Get the procedure

    Discovery, analysis, the change itself, verification and the exact shape the output must take — including what counts as evidence and what may not be asserted without it.

  5. Run it where your code already is

    Paste it into Claude Code or Codex inside your own environment. Nothing leaves your infrastructure. You decide afterwards what is safe to share.

  6. Bring the result back

    Upload what came out. It is checked for completeness, for missing evidence and for findings that contradict each other, and you get a suggested next step.

  7. Ask a person, if it matters

    A written expert assessment, or a recorded video walkthrough of your material. No meeting, no scheduling, no time zones.

Technical Context

The part that makes the second month better than the first

Everything the platform learns about your system is kept, and every action after that is written against it. That is the difference between a subscription and a one-off purchase.

What your system is made of

Stack, hosting, databases, deployment model, integrations, the AI coding tools your team actually uses, and the constraints you cannot change.

Who works on it

Your team, your contractors and your vendors — so an assessment of a proposal knows who wrote it and what they have promised before.

What you already decided

Past technical decisions and the reasoning behind them. New actions stop re-litigating settled questions and start from where you are.

What is still open

Unresolved risks and known weak spots, carried forward until they are closed rather than forgotten between conversations.

Your tools

Built for the tools you already run

The procedure is written to be executed by an AI coding agent inside your environment, with the evidence requirements a careful engineer would insist on.

Claude Code

Procedures are written in the shape Claude Code works best in: explicit discovery first, no modification during inspection, evidence cited by file and line.

Codex

The same procedure, adapted to how Codex approaches a repository — including using it as an independent second reader of work another model produced.

Anything else you use

The output format is specified, not the tool. If your team runs something else, the procedure still describes what must be produced and what counts as proof.

Why not just ask the model yourself

You can. Here is what changes.

Asking your AI directly

You start from a blank prompt every time, and the model starts from nothing it knew last week.

It answers confidently about a system it has only partly read, and you have no easy way to tell which parts it actually checked.

The output is shaped by how you happened to ask, so two audits of the same system are not comparable.

Nobody senior ever looks at the result unless you go and find someone.

Running a technical action

The procedure already carries your context and the decisions you made before, so it starts where you are.

It separates inspection from change, and requires evidence for every claim — a finding without a file and a line does not count.

The output has a fixed shape, so this month’s audit can be compared with the last one.

When it matters, a human expert reads the result and tells you which parts to believe.

When it does not go to plan

What happens when something fails

A failed generation costs nothing

Credits are held, not spent, until the action is actually produced. If it fails, the hold is released and you can retry.

An incomplete result is said so

If what you bring back is missing evidence or contradicts itself, you are told which part and why, rather than getting a confident summary of a half-finished job.

You are never left guessing the state

Every review shows where it is, whether anything is needed from you, and the service window it is running against.

Try it on the thing you have been putting off

Store your context once, describe the situation, and see the procedure before you spend a credit.

Choose a plan

Technical expertise on subscription. Personalised technical actions, AI-ready audits and independent expert review for founders, teams and software businesses.

Technical Actions

How it works

Who it is for

Legal

© 2026 TechnicalAngel.