Skyvern

Skyvern takes a task-based approach to browser automation — rather than scripting individual actions, agents submit a high-level task (URL + instruction + data), and Skyvern's LLM + vision system navigates the page and completes it. This makes it robust to page changes and suitable for automating web workflows on sites that change frequently or lack APIs. Hosted Skyvern Cloud or self-hosted. Confidence is docs-derived.

REST API at api.skyvern.com/v1. Resources: tasks. POST /tasks creates a task ({url, goal, data: {...}, proxy_location}). Returns task_id. GET /tasks/{id} polls task status (created/queued/running/complete/failed). GET /tasks/{id}/steps retrieves step-by-step execution log with screenshots. WebSocket: subscribe to real-time task events. Task execution is async; agents poll or use webhooks.

API key auth: x-api-key header. Keys from Skyvern Cloud dashboard. HTTPS enforced. No OAuth2. Self-hosted: no external API key required; manage locally. Credential storage for authenticated web sessions available (Skyvern stores credentials securely for reuse).

Task states: created → queued → running → complete/failed/terminated. Failed tasks include failure reason and last screenshot for debugging. Complex tasks may require multiple retries. Task timeout configurable. Skyvern Cloud uptime at status.skyvern.com. Self-hosted reliability is operator's responsibility.

docs.skyvern.com covers quickstart (cloud and self-hosted), task API reference, credential management, and step debugging. Getting started: create account, generate API key, submit first task in under 10 minutes. Open-source MIT license (GitHub: Skyvern-AI/skyvern). Community via Skyvern Discord. Free tier available.