Using References with Agents

Agents work better when you give them context. References are documents, files, and requirements you attach to help the agent understand what you need and how to deliver it.

How to use References

You might attach a product brief so the agent knows the features to highlight, a brand guide so it matches your tone, or a spreadsheet with data to analyze. The more relevant context you provide, the better the output.

Supported Formats

MeshMesh supports common document formats:

• Documents — .docx, .pdf, .txt, .md
• Spreadsheets — .xlsx, .csv
• Images — .png, .jpg, .gif

The agent reads and understands the content, using it to inform its work.

Adding References

You can add references when creating an agent from scratch, or anytime while you're working in an agent. There are three ways to do it:

Select existing references — Choose from documents you've used in other agents. Useful for standard materials like brand guides or process docs that apply across multiple projects.

Upload new files — Drag and drop files directly, or click to browse your computer. Uploaded files are saved and available for future agents.

Create a new document — Write or paste content directly in MeshMesh Studio. Helpful when you want to draft requirements on the fly or compile notes without leaving the app.

What to Include

Requirements and specs — What does the agent need to deliver? Include acceptance criteria, constraints, and must-haves.

Examples — Show the agent what good looks like. A sample email, a previous report, or a template to follow.

Data — Spreadsheets or CSVs the agent should analyze, reference, or use as input.

Brand and style guides — Tone of voice, terminology preferences, formatting rules.

Process documentation — How things work in your org, approval flows, naming conventions.

Best Practices

Break large requirements into phases — If you have a complex project with extensive requirements, don't try to do everything at once. Split it into subtasks or phases. Prompt the agent to focus on one phase at a time: "Plan the first phase" or "Build the first phase only." This gives you checkpoints along the way and makes it easier to course-correct.

Keep images lightweight — Large, high-resolution images slow things down and can hit size limits. Resize or compress images before uploading. If the agent just needs to understand what's in the image, a smaller version works fine.

Use accurate data — The agent trusts what you give it. If you provide sample data, make sure it reflects reality. Placeholder or fake data can lead the agent to make incorrect assumptions or produce outputs that don't match your actual use case.

Editing and Downloading References

References aren't static — they can evolve as your project progresses.

Manual editing — For text-based files like .md, .html, .json, and .txt, you can click on a reference to open it, make your changes inline, and save. The agent will use the updated version going forward.

Agent-assisted editing — For binary files like .docx and .xlsx, ask the agent to make changes for you. Just describe what you want updated and the agent will modify the file directly.

You can also ask the agent to create new references from scratch. If you're working through a project and want to capture decisions or document requirements, ask the agent to write them to a reference. This is a great way to build up project documentation as you go.

Downloading — Any reference can be downloaded to your computer. Click the download button on the reference to save a local copy.

Think of references as a shared workspace between you and the agent. The more you use them, the more context the agent has to work with.