Overview

Before you start clicking, here's the mental model. Mesh is built around four key concepts:

The Source Section

✕

Where your context comes from. Mesh supports five source types:

• Notion - pages and databases from your Notion workspace
• Local Files - code or text files from your filesystem, selected via glob patterns
• Context Blocks - your personal library of reusable prompts and snippets
• Git - diffs, commits, and changed files from a local Git repository
• GitHub - issues and pull requests from your GitHub repos

Each source has its own panel in the app. You configure each one independently.

The Stack Section

✕

Your assembly area. When you find an item from any source, a Notion page, a code file, or a GitHub issue, you can add it to The Stack. The Stack is where you compose the collection of reusable building blocks for LLM context generation.

You can drag items to reorder, click to remove, copy to clipboard, and edit content directly in the editor for each stack item. You can also reload the item from its source to revert changes or update to the latest version.

A model-aware token counter tells you how much of your selected LLM's context window you're using as you build, both for the entire stack and for each item.

Stacks can be saved and loaded from files so that you can create your own presets for common scenarios and context collections that are easy to recall.

The Output Section

✕

Your context output generation editor. In the output editor, you can:

• Combine stack items into a single context output
• Edit the context output with language-specific syntax highlights and nested collapsing
• Scan the context output for privacy issues
• Copy the context output to the clipboard or export it to a file
• Save and load context output presets from files

The Privacy Scanner

✕

Mesh also includes a built-in Privacy Scanner that checks for API keys, secrets, PII, and internal file paths. You can manually scan any content inside your stack or output sections, as well as automatically scan content when you copy it to the clipboard or export it.

• Detects privacy issues in the context based on built-in and customizable rules.
• Masks private data, while keeping the masked string consistent so the LLM can keep track of items that appear multiple times in the context.
• Auto masking, per-rule masking, and no masking for the entire content in a single click.
• All scanning is done locally. Your data is not being sent anywhere at any point in time when using Mesh, unless you explicitly copy it elsewhere.

By default, the privacy scanner is activated automatically when copying to the clipboard. If suspected private data is detected, a dialog will describe the findings and allow you to select which action you wish to take.

✕

You can also manually trigger the privacy scanner by pressing the scan shield icon inside the stack item editor and the output section menu.