Serif is an AI-native research workspace for LaTeX documents, providing a project-wide agent that can inspect files, verify sources, manage citations, revise sections, create figures, and compile the final PDF. Built for doctoral students and researchers, it handles multi-file projects with cloud compilation supporting pdfLaTeX, XeLaTeX, and LuaLaTeX. The platform includes a free tier with five daily agent trial messages and paid plans starting at $20 per month.
Funding
Funding not disclosed
Founders
Product
Problem
Researchers and technical authors working with LaTeX often rely on general-purpose AI chatbots, which require copying fragments of text out of context, losing project structure, and manually transferring edits, citations, and figures back into the document. This fragmented workflow is error-prone and inefficient, especially for complex, multi-file projects with bibliographies, figures, and custom class files.
Solution
Serif provides a project-wide AI research agent that operates directly within complete LaTeX documents. The agent can inspect project files, verify sources, manage citations, revise sections, create editable academic diagrams, generate original images, compile the document, and check the finished PDF. It works with the user's existing project structure, preserving document classes, packages, and folder organization. The agent runs tasks in the cloud, allowing long autonomous work to continue after the user leaves the page, and it can pause to ask for decisions when needed. All activity, edits, and compilation results remain visible and grouped by turn, giving users full control and the ability to revert changes at any level.
Target Audience
Primary users are doctoral students, researchers, research engineers, and technical authors producing complex, source-heavy LaTeX documents who need AI assistance integrated directly into their writing workflow.
Features
- Project-wide context awareness that reads the manuscript, bibliography, figures, included files, and compiler settings before applying changes
- Automatic build repair workflow that compiles after editing, inspects failures, fixes source code, and verifies the resulting PDF
- Citation verification with evidence inspection, preservation of existing bibliography entries, and visible source tracking
- Editable academic diagram creation and original image generation as project assets ready for compilation
- Cloud-based agent tasks with recoverable state, allowing long-running work to continue asynchronously
- Granular undo history grouped by Intelligence turn and file, supporting revert of individual changes, turns, or files
- Support for pdfLaTeX, XeLaTeX, and LuaLaTeX with automatic compiler selection based on project requirements
- Real-time collaboration with shared projects, live activity tracking, and feedback connected to specific work