Downloading Artifacts

Every run generates artifacts: video recordings, screenshots at each decision point, extracted data as JSON, and any downloaded files. Most are visible in the Run Details page, but you can also save them locally or pull them into another system.

From the UI

Artifact	How to download
Recordings	Recording tab → right-click the video → Save video as…
Screenshots	Overview tab → click a timeline item → right-click the image → Save image as…
Extracted data	Output tab → select the JSON text and copy
Downloaded files	Output tab → Agent Run Downloaded Files → click any link
Generated code	Code tab → Copy button in the top right

From the API

Use the API when you need artifacts programmatically: for a pipeline, an archive, or integration with another tool.

List artifacts for a run

curl -X GET "https://api.skyvern.com/v1/runs/{run_id}/artifacts" \
  -H "x-api-key: YOUR_API_KEY"

Filter by type using the artifact_type parameter (repeatable):

curl -X GET "https://api.skyvern.com/v1/runs/{run_id}/artifacts?artifact_type=screenshot_llm&artifact_type=recording" \
  -H "x-api-key: YOUR_API_KEY"

Download an artifact

Each artifact in the response includes a signed_url, a temporary, pre-authenticated download link:

{
  "artifact_id": "art_abc123",
  "artifact_type": "screenshot_llm",
  "signed_url": "https://storage.example.com/screenshot.png?token=...",
  "created_at": "2025-01-15T10:30:00Z"
}

curl -o screenshot.png "SIGNED_URL_HERE"

Signed URLs expire after a short period. Always generate a fresh URL via the API rather than storing old ones.

Artifact types

Value	What it is
`recording`	Full browser session video
`screenshot_llm`	Screenshots with numbered element annotations (what the AI analyzed)
`screenshot_action`	Screenshots taken after each browser action
`screenshot_final`	Final page state when the run ended
`llm_prompt`	Complete prompt sent to the LLM (goal, element tree, system instructions)
`llm_request`	Raw API request payload sent to the model provider
`llm_response`	Raw model response
`llm_response_parsed`	Parsed action list (what the AI decided to do)
`visible_elements_tree`	DOM tree of interactive elements the AI could see (JSON)
`html_scrape`	Full page HTML at the time of the step
`skyvern_log`	Formatted execution logs
`har`	HTTP Archive file (every network request the browser made)
`trace`	Browser trace file for performance analysis

The annotated screenshots (screenshot_llm) and parsed action lists (llm_response_parsed) are the most useful for debugging. Annotated screenshots show which elements the AI identified, and the action list shows what it decided to do with them.

Run Details

Understand what each tab shows and how to debug failures

Using Artifacts (API)

Full API reference for artifact retrieval

Getting Started

Building Agents

Agent Blocks

Running Tasks

Re-using Browsers

Viewing Results

Managing Credentials

Integrations

Account & Settings

Downloading Artifacts

From the UI

From the API

List artifacts for a run

Download an artifact

Artifact types

Run Details

Using Artifacts (API)

​From the UI

​From the API

​List artifacts for a run

​Download an artifact

​Artifact types

Run Details

Using Artifacts (API)

From the UI

From the API

List artifacts for a run

Download an artifact

Artifact types