TicketPipe — The IT ticket that remembers.
“ My VPN doesn’t work.” For IT teams, that sentence often starts a long exchange: Which error? Which application? What have you already tried? Employees struggle to explain technical problems, while support teams spend time reconstructing what happened.
TicketPipe uses Screenpipe’s recorded context to bridge that gap. Employees describe their problem in a few words. The application retrieves relevant recent screen activity and generates a structured ticket with observed error messages, application details, a timeline, attempted fixes, and a severity assessment based on the available evidence.
Before escalating, employees see suggested quick fixes. An optional “Let TicketPipe try” feature can attempt supported actions in a selected application window, with progress tracking and a Stop button.
If the quick fix works, no ticket is sent. If it does not, the employee can send the reviewed ticket to GitHub Issues, or save it locally when no integration is configured.
Sensitive information is masked locally before the context reaches the configured AI provider. Employees can inspect the context shared with the AI and review the ticket before sending it.
Our goal: less effort for employees, clearer evidence for IT, and fewer clarification loops.
TicketPipe is built with TypeScript and Bun, with an HTML/CSS/JavaScript interface served on localhost.
Screenpipe provides the incident context through its local REST API. TicketPipe queries recent OCR, accessibility, and input events, removes repeated text, groups activity by application and window, and highlights potential errors. A local redaction layer masks sensitive information before the context is sent to the configured LLM.
The AI returns structured JSON for ticket generation and troubleshooting suggestions. Supported integrations include Anthropic and OpenAI-compatible endpoints, including local model endpoints. Tickets can be published through the GitHub Issues API or saved as Markdown files.
Optional automatic repair uses Rust controllers and operating-system accessibility APIs to observe and interact with a user-selected window. The implementation includes action validation, bounded execution, cancellation, and independent result checks. macOS and Linux native desktop validation remains pending.
A bundled fictional incident and mock AI allow the core ticket workflow to run without Screenpipe or an API key.