Six steps from an empty project to an exported dataset. The same flow runs on the phone and in the web app — switch below to see the one you are using.
Everything in DeCiphr lives inside a project: its images, its framework, its coding, and its export. Start by creating one and naming it something you will recognize in your data six months from now.
Bring in the material you want to code — photographs taken in the field, or an image set you already have. Each image becomes a numbered item in the project, with its date and, where available, its GPS coordinates.
Images are stored on your device. An image is sent to Anthropic only when you run an analysis on it.
A framework is your coding scheme: the codes, their definitions, and the values each code can take. Pick one of your own, or import one from the public library.
DeCiphr runs in strict mode by default. The AI assigns values from the codes you defined and nothing else; where the image does not warrant a confident assignment it returns Uncodable rather than guessing.
Building your own? Read the Framework Guide →With a framework attached, run the coding. Each image is sent with your master prompt and your code definitions, and comes back with a value for every AI code in the scheme.
Items are coded one after another and marked as they finish.
Open any item to see what was assigned, code by code. Every value is editable. When you change one, your value is stored alongside the AI's original, so the record shows both what the model proposed and what you decided.
If a framework includes codes only you can answer — interpretive questions that need fieldwork presence — those stay blank for you to fill in. A framework that is coded entirely by the AI is equally valid; a second layer is optional, not required.
Export the project as an Excel workbook: one row per item, the AI's value and your edit side by side for every code, with the item number, coding date and framework name on every row.
Photographs export separately as a ZIP. In the mobile app, the files are numbered to match the rows, so the images and the coding stay traceable to each other.
The coding is only as good as the scheme behind it. The Framework Guide walks through master prompts, code definitions, controlled values, and how to decide what the AI should and should not be asked to answer.