docs(agent): add zero data retention details

api-reference/v2-openapi.json

@@ -1455,6 +1455,10 @@
                     "type": "number",
                     "description": "Maximum credits to spend on this agent task"
                   },
+                  "zeroDataRetention": {
+                    "type": "boolean",
+                    "description": "If true, request metadata is redacted and outputs are removed after the retention window"
+                  },
                   "strictConstrainToURLs": {
                     "type": "boolean",
                     "description": "If true, agent will only visit URLs provided in the urls array"

features/agent.mdx

@@ -1,7 +1,7 @@
 ---
 title: "Agent"
 description: "Gather data wherever it lives on the web."
 og:title: "Agent | Firecrawl"
 og:description: "Gather data wherever it lives on the web. Describe what you want, /agent handles the rest."
 sidebarTitle: "Agent"
 ---
@@ -45,7 +45,7 @@
 
 ## Using `/agent`
 
 The only required parameter is `prompt`. Simply describe what data you want to extract. For structured output, provide a JSON schema. The SDKs support Pydantic (Python) and Zod (Node) for type-safe schema definitions:
 
 <CodeGroup>
 
@@ -108,7 +108,7 @@
 
 ## Model Selection
 
 Firecrawl Agent offers two models. **Spark 1 Mini is 60% cheaper** and is the default — perfect for most use cases. Upgrade to Spark 1 Pro when you need maximum accuracy on complex tasks.
 
 | Model | Cost | Accuracy | Best For |
 |-------|------|----------|----------|
@@ -159,6 +159,37 @@
 | `model` | string | No | Model to use: `spark-1-mini` (default) or `spark-1-pro` |
 | `urls` | array | No | Optional list of URLs to focus the extraction |
 | `schema` | object | No | Optional JSON schema for structured output |
+| `zeroDataRetention` | boolean | No | Redact request metadata and schedule output cleanup after the retention window |
+
+## Zero Data Retention
+
+When `zeroDataRetention` is enabled, agent runs without storing activity logs or step traces. We keep the final output in GCS for up to 24 hours so status polling can succeed, then delete it when the ZDR cleanup job runs (after cleanup, `GET /v2/agent/:id` returns 404). ClickHouse activity and snapshot events are removed at completion.
+
+<CodeGroup>
+
+```python
+from firecrawl import FirecrawlApp
+
+app = FirecrawlApp(api_key="fc-YOUR_API_KEY")
+
+result = app.agent(
+    prompt="Extract the main features and benefits of Firecrawl",
+    zero_data_retention=True,
+)
+```
+
+```ts
+import Firecrawl from "@mendable/firecrawl-js";
+
+const firecrawl = new Firecrawl({ apiKey: "fc-YOUR_API_KEY" });
+
+const result = await firecrawl.agent({
+  prompt: "Extract the main features and benefits of Firecrawl",
+  zeroDataRetention: true,
+});
+```
+
+</CodeGroup>
 
 ## Agent vs Extract: What's Improved
 
@@ -185,7 +216,7 @@
 
 ## Pricing
 
 Firecrawl Agent uses **dynamic billing** that scales with the complexity of your data extraction request. You pay based on the actual work Agent performs, ensuring fair pricing whether you're extracting simple data points or complex structured information from multiple sources.
 
 ### How Agent pricing works