Frequently asked questions

Cryogram is a local tool that uses Claude Code or Codex to build automated workflows you can run at any time. You describe what you want, an AI agent figures out how to do it with you, then writes it into Python code. The workflow follows that process each time it runs, including any branches, and only calls AI for steps that need judgement.

For a one-off task, that may be all you need. For something you repeat, much of the work is usually the same: read these files, fetch these records, match them, update this system. Cryogram writes those parts into code once, so an agent does not need to work them out again on every run. If a step needs AI, such as extracting information from a document, it uses a specific prompt with a defined output, which can often be handled by a cheaper model.

Cryogram handles the parts around the code too: collecting inputs and secrets, installing dependencies, testing steps, checking their results, asking before sends, keeping run history and resuming after a problem. You can ask for changes in the same chat, review the plan and restore an earlier version if needed, without maintaining all of that yourself.

In Cryogram, you describe the process and the agent builds, configures and tests the steps for you. Integrations are not limited to a catalogue of pre-built connectors: Cryogram uses official documentation, skills and its own learnings to set up connections. You can see the resulting steps on a canvas, but you do not need to assemble the workflow there yourself.

No. You need to know what you want the workflow to do, have access to the systems involved and answer questions when something needs clarifying. Cryogram handles the code and setup, and gives you instructions when it needs something from you. You can inspect the code and prompts if you want to.

Repeated tasks such as checking a batch of expenses, updating a CRM from a spreadsheet, filing documents or turning meeting notes into actions against an existing project list. A workflow can ask for input, process lists and take different branches depending on what it finds. Its outputs can go into your files and existing tools; Cryogram builds workflows, rather than a new app interface for each task.

That lets workflows use local files and folders, internal systems and a browser where you can sign in to websites that are difficult to automate through an API. You do not need to set up a server or manage containers. Your computer needs to be on and Cryogram running while a workflow runs; closing the browser tab does not quit Cryogram.

Mac and Windows. It needs Python 3.10 or newer, and if your computer has none the installer offers to download a private copy into the Cryogram folder. The GitHub README has the install commands and instructions for installing from a source ZIP.

Yes. The AI agent that builds and changes workflows runs through Claude Code or Codex, using your existing subscription with no additional login required, or an API key if you prefer. Setup checks what is installed and guides you through what is needed.

You cannot use a subscription for AI steps within the workflows; these need an API key for the provider you choose. A local model may not need a key. Most workflows do not need any AI once they have been built.

Cryogram comes with presets for OpenAI, Anthropic and Gemini. You can add as many providers as you want, including multiple entries for the same provider with different keys for different purposes. Under Admin > Workflow AI providers, use + Add provider and choose OpenAI-compatible as the API type to connect to other compatible providers, such as xAI, Mistral, DeepSeek, Moonshot and OpenRouter, or a local model server.

When adding an AI step, Cryogram picks the cheapest suitable model with known pricing from your default provider. You can change the model or prompt by opening the step on the canvas.

No. There is no Cryogram account or sign-up. You use your own AI subscription or keys and your own accounts for any services the workflow connects to.

Cryogram is free to download and use, and its source is published under the GNU AGPL v3. Building uses your AI subscription or API key; AI steps use your chosen provider and model. Steps that do not use AI consume no AI tokens, though any services your workflow connects to may have their own charges.

The app runs locally, and a built workflow that only uses local files and code can run offline. Building through Claude Code or Codex, calling a hosted AI model or connecting to an online service needs an internet connection. Installation and downloading dependencies also need a connection.

Workflows, chat, settings, run history and saved results are stored on your computer, including data kept so a stopped run can continue. Workflows read and write to the files and systems you set up.

Anything used by the AI agent when building a workflow is sent to your AI provider, like any other project you use it for. Once a workflow has been built, the agent does not see its run data unless you choose to share it, for example when asking it to investigate a problem. An AI step sends its inputs to its configured model when it runs. The security page explains this in more detail.

Secrets such as API keys are stored encrypted, so the AI agent can see their names, but never their values; the workflow uses them when connecting to the services they are for. When asking you to provide secrets, it provides a box for you to enter them into rather than the chat. Use that box rather than pasting a secret into a message.

During a build, Cryogram asks before connecting to a new address, accessing a folder you have not allowed, opening a browser window or sending something for real. A built workflow asks before steps that send, post or write to another app, unless you turn that approval off for the step. Opening a website can still leave a record of your visit, even when the step is only reading it.

Cryogram tests steps while building and adds checks that run with the workflow. If you decline a live test, such as sending an email or opening a browser, the step can still be built, but your first run is its first real attempt. Cryogram is in beta, so test your workflows a few times before relying on them blindly.

Its checks catch problems and stop the run, keeping a snapshot so you can continue after fixing it. Open Issues and choose Investigate to ask the agent to look into it, review its proposed fix, then use Continue once it is ready. If a send was not confirmed, check whether it arrived before repeating it.

Yes. Describe the change in its chat and review Here's what will change before approving it. You can see previous versions under Information > Version history and choose Restore to return to one. Restoring keeps your current saved values, chat and run history.

Yes. Click a step on the canvas to open its details. The workflow files are also stored in your application data folder, and Cryogram's own source is on GitHub.

Information contains instructions for the workflow, and Knowledge explains how to use the app. Help & feedback opens the support page. You can also open a GitHub issue or write to hello@cryogram.app. If Cryogram offers a problem report after a failed run, you can review it before deciding whether to send it.