> ## Documentation Index
> Fetch the complete documentation index at: https://lava.so/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Importing Workflows

> Bring the skills and instructions you already use in Claude, ChatGPT, Claude Code, and other AI tools into Lava as Workflows.

You may already have instructions that work: a Claude Code skill that posts your standup, a custom GPT that screens candidates, a Claude project set up for sales follow-ups. Import brings them into Lava. Lava drafts a Workflow from each one in a conversation with you, and nothing is saved until you approve it.

<Info>
  **Prerequisites:** Lava Desktop installed and signed in. See [Install Lava Desktop](/docs/desktop/install) if you haven't done this yet.
</Info>

## How Importing Works

<Steps>
  <Step title="Open Import">
    On Home, open **Workflows**, then choose **Import from other tools**. You can also start from **Import from other tools** on Home when you have no Workflows yet.
  </Step>

  <Step title="Pick where your instructions live">
    The Import page lists any skills Lava found on this computer, then **Claude**, **ChatGPT**, and **Another tool**, then files. Each source opens its own steps.
  </Step>

  <Step title="Choose what to bring in">
    Tick the skills or projects you want, or paste the instructions. A line above **Bring in** says what stays on your computer.
  </Step>

  <Step title="Review in the conversation">
    **Bring in** starts a new conversation with your files attached. Lava drafts a Workflow from each one, shows you what it does and what it needs, and asks before saving each one.
  </Step>
</Steps>

Each import comes from one source. To bring in skills from Claude Code and a custom GPT, import each in turn; each opens its own conversation.

## What Stays on Your Computer

Nothing leaves your computer until you choose **Bring in**. To list the skills it found, Lava reads only their names and descriptions on this computer. Then:

* **Scripts and programs stay behind.** Lava never runs imported code. The Workflow describes the script's job in words, and Lava does that job with your connected apps.
* **Secrets stay behind.** Files such as `.env`, `.pem`, `.key`, SSH keys, `credentials*.json`, `.npmrc`, and `.netrc` are skipped, and so is any text that contains an API key or token.
* **Your conversations stay behind.** From a Claude export, Lava takes only each project's instructions and Claude's memory. Conversations, account details, and project knowledge files stay on your computer.

Select **Details** beside **Bring in** to see each file that stays behind and why. What you bring in goes to that conversation's workspace in Lava.

## Import from This Computer

When Lava finds skills that other AI tools keep on this computer, they lead the Import page, one row per tool. Open a tool to see its skills and tick the ones you want. If a tool keeps five skills or fewer, they arrive ticked; a longer list starts unticked, with a filter once it passes eight. Skills already in Lava are marked **Already in Lava** and stay unticked.

Lava checks these folders:

| Tool | Folder |
| - | - |
| Claude Code skills, including skills synced from claude.ai | `~/.claude/skills` |
| Claude scheduled tasks | `~/.claude/scheduled-tasks` |
| Claude Code commands | `~/.claude/commands` |
| Codex, Cursor, or Gemini CLI | `~/.agents/skills` |
| Cursor | `~/.cursor/skills` |
| Gemini CLI | `~/.gemini/skills` |

Skills kept anywhere else come in from files: drop the folder on the page or choose it.

<Note>
  A Claude scheduled task holds its prompt but not its schedule. Lava asks when it ran, and offers to set it up as a Routine after you try it.
</Note>

## Import from Claude

Claude includes your projects and memory in your data export. Lava reads the export on this computer and brings in only what a Workflow or your Wiki can use.

<Steps>
  <Step title="Request your data export">
    On the Import page, choose **Claude**, then **Open Claude settings**. In Claude, open **Settings**, then **Privacy**, and choose **Export data**.
  </Step>

  <Step title="Download the .zip">
    Claude emails you a download link, which expires after 24 hours. Download the `.zip` from it. If the link expires, request the export again.
  </Step>

  <Step title="Bring it here">
    Drop the `.zip` on the page, or choose **Choose the export**. Lava lists the projects and memory it found.
  </Step>

  <Step title="Tick what you want">
    Untick anything you'd rather leave, then choose **Bring in**.
  </Step>
</Steps>

**What comes in:** each project's custom instructions, which become Workflows, and Claude's memory, marked **Suggested for your Wiki**.

**What stays behind:** your conversations, your account details, project knowledge files, and projects with no instructions.

<Note>
  Claude offers data exports on its Free, Pro, and Max plans. On a Team or Enterprise plan, your organization's Primary Owner requests the export. You can still paste a project's instructions into **Another tool**.
</Note>

## Import from ChatGPT

Lava brings in a custom GPT from its instructions, which you copy from ChatGPT.

<Steps>
  <Step title="Open your GPTs">
    On the Import page, choose **ChatGPT**, then **Open ChatGPT**. Your GPTs open in Lava's browser. When you come back to Home, the Import page is waiting where you left it.
  </Step>

  <Step title="Copy the GPT's instructions">
    Choose the GPT, select **Edit GPT**, then **Configure**, and copy everything under **Instructions**.
  </Step>

  <Step title="Paste and bring in">
    Paste the instructions into the box and choose **Bring in**.
  </Step>
</Steps>

**What comes in:** the GPT's instructions, which become one Workflow.

**What doesn't:** anything you don't paste. The GPT's knowledge files, actions, and conversation starters stay in ChatGPT.

## Import from Another Tool

Choose **Another tool** to paste instructions from anywhere: a Gemini Gem, a Copilot agent, a saved prompt, or a page of steps from your team's wiki. Lava drafts a Workflow from what you paste.

## Import from Files

Drop files anywhere on the Import page, or choose **choose a folder** or **choose files**. Lava lists what it found as recipes, such as a skill folder or a Markdown file, for you to tick. You can bring:

* **Skill folders:** a `SKILL.md` with its references and templates.
* **Markdown and text files:** `.md`, `.txt`, `.json`, `.yaml`, `.csv`, and `.html`.
* **A `.zip`** of either, up to 10 MB.
* **A Claude data export**, which Lava reads as described in [Import from Claude](#import-from-claude).

Each file can be up to 1 MB, and one import holds up to 200 files and 10 MB. Scripts, programs, secrets, and other file types stay on your computer.

## After You Bring Something In

Lava works through your import in the conversation:

1. It reads each file and sorts it: a recipe becomes a Workflow, and standing guidance such as a CLAUDE.md or Claude's memory becomes a suggestion for your Wiki.
2. It shows each draft in plain words: what it does, the apps it needs and whether they're connected, whether it sends anything, and what didn't carry over.
3. It asks about each draft: save it, change something first, or skip it. Nothing is saved until you choose to save.
4. After saving, it offers to try the Workflow now. If the original ran on a schedule, it offers to try it and then run it on that schedule as a Routine. Lava asks for your approval before creating a Routine, unless the agent is in Auto mode.

Each imported Workflow ends with a line that names where it came from, such as `Imported from Claude skill "daily-standup" on 2024-01-01.` Lava uses that line to notice when you import the same recipe again, and asks before making a second copy.

Lava suggests what could go in your Wiki but writes nothing there. You add a suggestion yourself.

## FAQ

<AccordionGroup>
  <Accordion title="Does anything leave my computer before I choose Bring in?">
    No. Lava reads the names and descriptions of found skills on this computer to list them. Files go to Lava only when you choose **Bring in**, and only the ones you ticked.
  </Accordion>

  <Accordion title="Does Lava sign in to Claude or ChatGPT for me?">
    No. You request the Claude export and copy a GPT's instructions yourself. Lava opens those pages in its browser so you don't have to find them, and never acts in your Claude or ChatGPT account.
  </Accordion>

  <Accordion title="Can I import a skill that runs a script?">
    Yes, without the script. Scripts stay on your computer and Lava never runs imported code. The Workflow describes what the script did, and Lava does it through your connected apps. If an app it needs isn't connected, Lava tells you which one.
  </Accordion>

  <Accordion title="Why doesn't Lava find my skills?">
    Lava checks only the folders listed under [Import from this computer](#import-from-this-computer). If your skills live somewhere else, drop the folder on the Import page or use **choose a folder**.
  </Accordion>

  <Accordion title="Does importing change the original skill, project, or GPT?">
    No. Lava reads a copy. Your original stays as it was in the tool it came from.
  </Accordion>

  <Accordion title="Can I import a Zap or another automation?">
    Not directly yet. Paste a description of what it does into **Another tool**, and Lava drafts a Workflow from your description.
  </Accordion>

  <Accordion title="What happens if I import the same skill twice?">
    Lava marks skills that are already in Lava and leaves them unticked. If one comes through anyway, Lava notices the earlier copy and asks before saving a second.
  </Accordion>
</AccordionGroup>

## Next Steps

<CardGroup cols={2}>
  <Card title="Working with the Agent" icon="message" href="/docs/desktop/working-with-agent">
    Run your Workflows and follow the agent's work.
  </Card>

  <Card title="Using Lava Safely" icon="shield-check" href="/docs/desktop/using-lava-safely">
    How approvals keep you in control of what the agent does.
  </Card>
</CardGroup>


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.