A method for you & your AI agents

Build with AI.
Keep your
direction.

Turn an idea into a clear mission. Make the choices that matter, let your agents build, and keep the evidence behind the result.

Open source MIT licensed Built for real work

THE THREAD OF A MISSIONIllustration
01 — YOUR INTENT

A small idea.
A clear outcome.

02 — A CHOICE THAT STAYS

Direction, decided.

03–05 — BUILD. VERIFY. RESUME.

The work.
And the why behind it.

DecisionsEvidenceNext step
One continuous thread. Even when the conversation changes.

Bring the agent
you work with.

CodexClaude CodeCursor Agent
Installation profiles
& compatibility

01 / THE METHOD

One mission.
A thread worth keeping.

A mission names the result you want, its boundaries, and how you will check it. Use the steps that help. A small fix can stay small.

FIRST, MAKE THE NEED CONCRETE.

Know what
good looks like.

Give the agent an outcome, useful constraints and a way to tell whether the work meets the need.

What you keepA mission with a clear finish line.
MISSION / READING LIBRARY

Find the book
you want to read next.

The need
Filter a reading list by status.
The boundary
Keep the existing design and books.
Success looks like
All, In progress and Finished show the matching books.

KEEP THE CHOICES THAT MATTER.

A direction.
Not a guessing game.

Examine the product, visual or technical choices that shape the work. Reuse decisions you have already accepted.

What you keepA decision and its reason.
DECISION / INTERACTION

Keep the filters
in plain sight.

Visible status filtersChosen in this example
A dropdown menuAlternative considered

Three short choices fit beside the list. Readers can see which one is active.

LET THE AGENT MOVE THE WORK FORWARD.

Build something
you can try.

Hand over a concrete scope, the relevant references and clear responsibilities. Inspect the result as it takes shape.

Try the exampleChoose a status to filter this little library.
INTERACTIVE EXAMPLE

A little reading list.

  • The quiet daysIn progress
  • An atlas of pathsFinished
  • The house by the tidesIn progress

LOOK AT THE WORK. THEN AT THE EVIDENCE.

“Done” needs
something behind it.

Review the changes, run relevant checks and try the important journeys. Keep failures and unverified behavior visible.

What you keepResults tied to the version you examined.
VERIFICATION / PLANNED CHECKS

Check the behavior.
Not just the build.

  1. 01
    Try each filter

    Only matching books should remain.

  2. 02
    Use a keyboard and a small screen

    The controls should stay reachable.

  3. 03
    Record the actual results

    Link the evidence to this version.

LEAVE THE NEXT STEP IN REACH.

Pick up the work.
Keep the context.

A useful handoff preserves the decisions, current files and remaining checks. Resume from what is real, even in a new conversation.

What you keepA clear place to start again.
HANDOFF / EXAMPLE NOTE

For the next
pair of hands.

Keep
The accepted visible-filter design.
Inspect
The current implementation and recorded checks.
Continue
Resolve remaining failures before considering the mission complete.
YOU

Set the direction.

Define the outcome, constraints and how much to delegate. Make the important calls.

YOUR AGENTS

Move the work forward.

Carry out authorized work, surface uncertainty and prepare results you can examine.

THE EVIDENCE

Make progress legible.

Show what was checked, what changed and what still needs attention.

Read how missions work

02 / DEVMethod STUDIO

The project.
The choices.
The whole picture.

In development · Not yet released

Studio brings the brief, conversation, visual references, code versions and their checks into one local workspace.

Explore the Studio guide
01

Start or pick up

Describe an idea, return to a project or import a copy of existing work.

02

Try and adjust

Open the local preview, inspect the files and prepare the next change.

03

Keep versions clear

Distinguish active code, drafts and the checks associated with each version.

Studio is in development and has not yet been published on GitHub. The capabilities described here are a preview and may change before release. It relies on a host agent and supports React / TypeScript and HTML / CSS / JavaScript. Local preview and export are documented; public hosting and application authentication require their own setup.

A REAL LOCAL PROJECT / 1 MIN 28

See the work.
Follow the decisions.

This video shows an earlier development version of Studio. The interface and workflow have since changed; an updated recording is in preparation. Open the Studio video page.

Edited from real local Studio captures. Synthetic English narration and English captions; the recorded interface is in French. Earlier development version; not the current interface.Download the film
Read the English transcript

This is DevMethod Studio: a local workspace that connects your idea, your coding agent’s work, and a result you can actually try.

Here, we’re building the DevMethod website. The brief captures its audience, content, and visual direction. The decisions that shape the work stay with the project.

Guides help prepare the tools you need. An MCP server can support your assistant across projects. An application API serves the product you’re building. Preparation and a verified connection remain distinct.

Once the agent has produced a version, the preview shows the real application. Inspect the result, open its source, and prepare the next change from an element on the page.

Each check keeps its result and the version it examined. A passing result describes what was tested. Failures and missing checks stay visible, so you can decide what needs attention.

Testing exposed broken documentation links and a small-screen layout issue. We asked for the fixes in Studio, then checked the next version. Earlier results stay attached to their original version.

The project can be exported with its code and context for local use. DevMethod Studio: build with your agents, with decisions and evidence you can inspect.

CONNECTED, WITH CONTEXT

The right access.
For the right purpose.

A connection for your agent and an integration for your application have different jobs.

Read the connection boundaries
A

Tools for the agent

Workspace MCP connections can be reused and selected per project. The local bridge applies tool permissions: allow, ask or deny.

For example: project context in Notion.
B

Services for the application

Project API choices express what your app needs. The adapter, runtime credentials and working integration still need to be built and verified.

For example: payments inside your product.
C

Accounts for your users

Your own connection does not become your users’ connection. Their sign-in, consent and access belong to the application’s design.

Identity stays explicit.

03 / FROM THE REPOSITORY

From a first choice
to a working library.

Meet Lisière: a fictional reading library, and a real recorded example of the method in use.

The walkthrough compares visual directions, keeps a reference, then builds a small application with forms, filters and reading statuses. The sources include tests, desktop and mobile browser checks, and the corrections made along the way.

Watch the walkthrough Recorded in French · 4 min 03Explore the source & evidence
LISIÈRE / A VISUAL PILOT
The Lisière reading-library example: three illustrated book covers, status filters and controls to add books or update reading progress.
Actual desktop capture from the repository’s visual pilot. The library and its books are fictional. The film also includes a review-interface chapter using labelled fictional data.

04 / YOUR FIRST MISSION

Start with
one concrete need.

Bring a small fix or a focused feature. Keep your repository’s conventions. Give your agent a result it can work toward.

Open the getting-started guide
A FIRST REQUEST COULD BE
Add a status filter to this list. Keep the existing design, verify the behavior, and leave a clear point to resume from.
Small scope. Observable result.

01 Choose your agent profile.02 Review the proposed setup files.03 Describe your first mission.

The CLI requires Node.js 22+ and npm. Tool availability and command discovery depend on your agent. See compatibility

A FEW GOOD QUESTIONS

Clear scope.
Open answers.

The method, code, examples and their limits are there to examine.

Explore the repository
Do I need to follow every step?

No. Use the entry point that fits the work: framing, implementation, review or verification. A simple correction does not need the same process as a new product.

Does DevMethod work with my agent?

Installation profiles are provided for Codex, Claude Code and Cursor Agent. Installation tests and real agent trials cover different things. The compatibility document explains what has been checked.

Does DevMethod generate the code and images?

The method guides your agent. The host agent performs the work, and image generation needs an image tool available in that host. Studio also relies on a host agent to carry out requests.

Is Studio a hosted app builder?

Studio is in development and is not yet available on GitHub. The video shows an earlier version. The documented runtime offers local preview and export. It does not provide automatic public deployment or application authentication on its own.

Does selecting a connector connect my application?

No. A project service choice prepares an integration intention. Workspace MCP connections have a separate connection and permissions flow for the agent. Application access, runtime credentials and end-user accounts must be implemented for the project.

Can I use and contribute to DevMethod?

Yes. DevMethod is open source under the MIT license. The contribution guide explains how to work with the repository.