Work with your agent

Your agent can work on the same Tylor project you open in your browser. It can fetch source files, use your templates and fonts, propose edits, compile a PDF, inspect rendered pages, and create a History checkpoint. You do not need to copy each result from a chat response back into the editor.

Connect your agent to get started.

Codex, Claude Code, Pi, and OpenCode can use this workflow. Other agents need terminal access and a way to load the Tylor skill.

How the pieces fit together

Piece	Role
Your agent	Interprets your request and carries out the task
Tylor skill	Gives the agent instructions for operating Tylor
Tylor CLI	Reads and changes projects through the Tylor API
Tylor browser editor	Shows source, preview, comments, suggestions, and History

The skill is an instruction file installed in your agent. It uses your local CLI and Tylor credentials; it does not require a separate Tylor MCP server. Your agent needs permission to run the CLI and reach your Tylor API from its execution environment.

Give the agent a project and a task

Start with the project name or ID and the operation you want. For example:

Use the Tylor skill. Find “ACM paper · Shared evidence” and
check why the table in methods.typ overflows. Fix the layout
without changing its values or wording. Compile and inspect
the PDF, then tell me which files changed.

Or ask for reviewable changes:

Use the Tylor skill on “Tylor · Paper review meeting”.
Check the selected wording for typos and propose tracked changes.
Keep the decisions, names, dates, and action owners unchanged.
Leave the suggestions pending for me to review in Tylor.

See the CLI in action

This terminal preview shows the commands behind a simple agent task: find a project, compile it, and render its pages for inspection. Use Step through and Next step to follow the sequence, or copy a command to your own terminal.

Find a project and check its PDFIllustrative session

$ tylor list

9e12a4f07b634f6283cc49a1c20d6e85  ACM paper · Shared evidence

$ tylor export 9e12a4f07b634f6283cc49a1c20d6e85 paper.pdf

compiled → paper.pdf

$ tylor render 9e12a4f07b634f6283cc49a1c20d6e85 --contact --out pages.png

pages.png  (contact sheet — 2 pages)

Sample output. Commands are not executed on this page.

The project ID and output are fictional. Use an ID returned by your own tylor list and choose local output paths for your files. An agent can run these commands for you, then inspect the rendered pages when its tools support images.

Follow the work in Tylor

Open the project in the browser to inspect the updated source and preview. Comments stay attached to passages; tracked changes can be accepted or rejected. Direct file updates are different from suggestions: ask for the mode you want before the agent starts.

If you also edit in the browser, wait for your edits to sync before the agent pulls files. Keep concurrent work on separate files when possible. Review agent changes covers the handoff, conflicts, and checkpoints.

Work beyond the active file

The CLI also gives the agent access to project and library operations:

• Start a project from one of your templates and place it in a folder.
• Upload source files, figures, or fonts you provide.
• Read project comments and inspect pending suggestions.
• Export a PDF and render pages to check the result.
• Create a History checkpoint or freeze a submission snapshot when requested.

An agent operates with the permissions of your API key. Your agent settings control its tool access and approvals. Content it reads enters that agent’s context.

External agents and the built-in AI tools

Your agent uses the skill and CLI from your agent environment. Tylor’s Assist and Ask tools run from the editor and use the instance’s configured AI service. Connecting an external agent does not require enabling the editor’s AI service; access to the agent is configured separately.

Continue with skills and reusable workflows if you want the agent to apply your own recurring instructions.