After spending the last three weeks meticulously tuning a complex procurement agent with a custom Python toolchain and a very specific prompt hierarchy, I had a cold realization: Lindy's UI is great for building, but I couldn't find a clear, automated export path. My entire configuration was living in their cloud, bound to my account session. This is an unacceptable single point of failure for any serious workflow.
The immediate, manual solution is to use the browser's developer tools, but it's tedious. Here's the process I've been using:
* **Navigate** to your agent in the Lindy web app.
* **Open DevTools (F12)**, go to the **Network** tab.
* **Refresh** the page. Look for a fetch/XHR request with a name like `agent` or `config`. The one you want usually contains the full JSON configuration.
* **Right-click** that request → **Copy** → **Copy as cURL**.
* **Run** that cURL command in a terminal, piping the output to a file. You'll need to preserve your session cookie.
```bash
curl 'https://api.uselindy.com/api/agent/your-agent-id-here'
-H 'Authorization: Bearer YOUR_TOKEN_HERE'
-H 'Cookie: YOUR_SESSION_COOKIE_HERE'
--compressed > lindy_agent_backup_$(date +%Y%m%d).json
```
This gets you a JSON dump, but it's a manual, per-agent process. The real questions for the community and Lindy are:
* Is there a **official, documented API endpoint** to fetch a full agent configuration (including tool definitions, prompt versions, and settings) programmatically?
* Does anyone have a **simple script** that authenticates and iterates through all agents/assistants to back them up locally?
* What's the **recovery story**? If I have this JSON blob, can I re-import it to recreate the agent identically, or is it just for reference?
Without a systematic backup and restore capability, building anything non-trivial on the platform introduces significant operational risk. The time investment in tuning prompts and chains is the real value; losing that to an account glitch or a UI misclick would be a major setback. I'm looking for a robust, scriptable solution.
Show me the benchmarks