# Cyberdesk

Cyberdesk is an AI-powered desktop automation platform built for real-world software operations. It helps teams automate repetitive and high-stakes work across Windows desktops and virtual machines, including legacy applications, EHR systems, ERP interfaces, and browser-based workflows. Cyberdesk combines natural-language workflow authoring, deterministic replay through trajectory caching, and production-ready APIs so organizations can run desktop automation reliably, repeatedly, and at scale.

## What We Do

Cyberdesk turns complex desktop procedures into reusable workflows.

Teams create workflows in the Cyberdesk Dashboard using clear instructions, optional prompt images, and optional output schemas. They then execute those workflows as runs on connected machines. A successful run can become a reusable trajectory (a recorded path of actions), which can be manually approved and replayed on future runs for dramatically faster execution.

The platform is built to handle both deterministic replay and dynamic runtime behavior. Even in cached workflows, dynamic tools like `focused_action`, `extract_prompt`, and `copy_to_clipboard` can still capture fresh, current data.

## Key Benefits and Capabilities

### Automate Real Desktop Work
Cyberdesk is designed for software that traditional browser automation often cannot handle well: desktop apps, thick clients, virtual desktops, and mixed environments. This makes it useful for operations that depend on legacy UIs, enterprise systems, and visual workflows.

### Get Faster Over Time with Trajectories
Cyberdesk records successful workflow executions as trajectories. After manual approval, those trajectories can be replayed in seconds instead of rerun from scratch. This provides substantial speed and consistency benefits while still allowing dynamic checks and extraction where needed.

### Keep Dynamic Logic in Cached Runs
Cached replay does not mean static behavior. Cyberdesk supports dynamic observations and decisions through:
- `focused_action` for context-aware decisions and extraction
- `extract_prompt` for vision-based extraction (sync, batch-scoped async, or run-scoped async)
- `copy_to_clipboard` for deterministic clipboard capture

This combination enables both performance and adaptability.

### Build Structured Outputs from Unstructured Interfaces
Workflows can define output schemas, and Cyberdesk transforms captured observations and runtime values into structured `output_data`. This makes desktop automation outputs easier to consume by downstream systems, APIs, and analytics pipelines.

### Run Multi-Step Pipelines on the Same Machine
With sessions and chains, teams can reserve a machine and execute multiple workflows back-to-back with shared desktop state. This is useful for workflows that require persistent login context, temporary files, and strict sequence control.

### Scale Programmatically
Cyberdesk provides:
- A REST API at `https://api.cyberdesk.io`
- Official TypeScript and Python SDKs
- Webhooks (Svix-backed) for event-driven architecture
- Native integration patterns for Slack and Zapier

This allows teams to embed Cyberdesk into existing products, backend services, and orchestration systems.

## Platform and Runtime Model

Cyberdesk is organized around a few core concepts:

- **Workflows**: Prompt-based automation definitions with optional schema, images, and variables
- **Runs**: Executions of workflows on a machine
- **Machines**: Connected desktops/VMs powered by Cyberdriver
- **Sessions**: Exclusive reservation of a machine for sequenced execution
- **Trajectories**: Recorded and approved action paths for deterministic replay

### Variable System
Cyberdesk supports three variable types in prompts:
- `{variable}` for standard inputs
- `{{variable}}` for runtime values discovered during execution
- `{$variable}` for sensitive values handled via secure vault aliasing

This lets teams separate ordinary parameters, runtime discoveries, and secrets with clear semantics.

## Cyberdriver: Desktop Connectivity Layer

Cyberdriver is the executable that connects machines to Cyberdesk Cloud and enables remote automation.

Key Cyberdriver capabilities include:
- Stealth mode on Windows (background, invisible process by default)
- Keepalive mode to reduce idle lock disruptions
- Black screen recovery for RDP/console session issues
- Persistent virtual display support for resilient unattended automation
- Remote updates from the dashboard
- Coordinate capture and log tooling for diagnostics
- Optional custom-host proxy routing for whitelisting/compliance/privacy requirements

Cyberdriver uses an outbound reverse tunnel model, which avoids opening inbound firewall ports in typical setups.

## Security, Privacy, and Control

Cyberdesk documentation explicitly emphasizes controls around sensitive data and operational security.

### Sensitive Inputs Are Vaulted and Ephemeral
Sensitive variables (`{$...}`) are:
- Stored in a secure third-party vault (Basis Theory) for run duration
- Not logged in Cyberdesk
- Not sent to LLMs
- Only resolved at execution time for actual computer actions (for example, typing)
- Deleted after run completion

Webhook payloads include sensitive input aliases (keys), not plaintext values.

### Webhook Authenticity and Reliability
Cyberdesk webhooks use Svix signatures and support standard verification headers (`svix-id`, `svix-timestamp`, `svix-signature`). Deliveries include retries with exponential backoff and support idempotent processing using message IDs.

### Operator-Controlled Guardrails
Teams can:
- Manually approve trajectories before replay
- Edit trajectory steps and cache detection behavior
- Use custom cache detection prompts for strict step-level validation logic
- Configure org-level run data retention rules

## Pricing and Billing Model

Cyberdesk uses usage-based billing with a step-oriented model.

- **Agentic steps**: AI decision/execution steps
- **Cached steps**: Replay steps from approved trajectories (faster and cheaper)

Billing depends on run outcome:
- Billed: success, user task failed, cancelled
- Not billed: infrastructure failure

This aligns cost with real work performed while protecting customers from infrastructure-side failures.

## Who Cyberdesk Serves

Cyberdesk is built for teams that need reliable automation across desktop-heavy operations, including:

- Engineering teams embedding desktop automation into products
- Operations teams running high-volume back-office workflows
- Healthcare and clinical operations handling EHR workflows
- Finance and revenue operations teams handling ERP/reporting tasks
- Organizations with legacy systems that cannot be automated through browser-only tools

## Common Use Cases

Cyberdesk documentation and examples highlight patterns such as:

- EHR workflows (for example, Epic-related extraction and updates)
- Patient data extraction and structured output generation
- Financial reporting and spreadsheet-based transformations
- Claims and document processing flows
- Order and invoice extraction from non-copyable interfaces
- Multi-step login -> process -> export workflows using sessions/chains

## Integrations and Ecosystem

Cyberdesk fits into broader automation stacks through:

- API + SDK integrations (TypeScript and Python)
- Event-driven webhooks (`run_complete` and cataloged event types)
- Slack notifications and custom Block Kit transformations
- Zapier connectivity to thousands of downstream apps
- Webhook transformation support for payload shaping and size reduction

## Getting Started

Typical path to value:

1. Create a workflow in the dashboard
2. Install and connect Cyberdriver on a Windows machine
3. Execute runs via dashboard or SDK
4. Review runs and approve successful trajectories
5. Add webhooks for downstream automation
6. Optimize throughput with async extraction and caching

## Resources

- Docs homepage: `/`
- Quickstart: `/quickstart`
- Cyberdriver setup: `/cyberdriver/quickstart`
- Blog: `https://cyberdesk.io/blog`
- Blog RSS feed: `https://cyberdesk.io/blog/feed.xml`
- Changelog: `https://cyberdesk.io/changelog`
- Blog post: `https://cyberdesk.io/blog/desktop-automation-runtime`
- Blog post: `https://cyberdesk.io/blog/trajectory-replay`
- Blog post: `https://cyberdesk.io/blog/ehr-automation-sessions`
- Blog post: `https://cyberdesk.io/blog/claims-processing-desktop-agent`
- Blog post: `https://cyberdesk.io/blog/erp-reporting-with-ai-desktops`
- Blog post: `https://cyberdesk.io/blog/sensitive-inputs-desktop-automation`
- Blog post: `https://cyberdesk.io/blog/ap-invoice-processing-desktop-automation`
- Blog post: `https://cyberdesk.io/blog/mortgage-loan-document-review-desktop-automation`
- TypeScript SDK guide: `/sdk-guides/typescript`
- Python SDK guide: `/sdk-guides/python`
- Webhooks quickstart: `/webhooks/quickstart`
- API intro: `/api-reference/introduction`
- OpenAPI spec: `https://api.cyberdesk.io/openapi.json`
- Dashboard: `https://cyberdesk.io/dashboard`
- Community: `https://discord.gg/ws5ddx5yZ8`

Cyberdesk helps teams move from brittle, one-off desktop scripts to durable, observable, and scalable automation systems. By combining trajectory-based speed, dynamic runtime intelligence, and integration-friendly APIs, the platform enables organizations to automate complex desktop work with higher reliability and lower operational overhead.