Skip to main content

​
Overview

Context allows you to attach files directly to your agent’s context window. Every file you add here is injected in full into the agent’s context on every run — no search, no retrieval, just the raw file content available to the model at all times.

Context

Files injected directly and fully into the agent’s context window on every run.

Knowledge

Files stored in a searchable knowledge base. Relevant chunks are retrieved at runtime using RAG.
Context is currently in preview and gated behind the ContextFiles feature flag. Contact support if you don’t see the Context section in your agent builder.
Use Context when:
  • The file is small enough to fit within your model’s context window
  • The agent needs to reference the entire file, not just relevant sections
  • You want a simple setup without configuring a separate knowledge base
Use Knowledge when:
  • You have large documents where only parts are relevant to each query
  • You need the agent to search across many files at once
  • You want semantic search over a large dataset

​
How to access Context

  1. Open your agent in the Builder
  2. Click Context in the left sidebar

​
File type viewers

Each file type has an optimized viewer so you can inspect what the agent will see before it runs.

​
Markdown files

Markdown files are rendered at reading width. A collapsible Properties section at the top of the viewer shows YAML frontmatter when the file includes it, so you can confirm metadata like title, tags, or date without scrolling the raw file.

​
CSV and TSV files

CSV and TSV files are displayed as scrollable tables with row numbers. For large files, a truncation notice appears to indicate that only a portion of the file is shown in the preview — the full file is still injected into the agent’s context.

​
Code files

Code files are displayed with line numbers and syntax highlighting based on the file extension.

​
Images

Images are displayed in a framed viewer.

​
Binary, Excel, and unknown files

Files that cannot be previewed — including binary files, Excel workbooks (.xlsx, .xls), and unrecognized file types — show a fallback card explaining that a preview is not available. The card includes an Open or Download button so you can still access the file.
Excel files cannot be rendered directly. To use spreadsheet data in an agent, convert the file to CSV first, or use the Convert a spreadsheet to JSON tool step in a tool.

​
Empty files

If a file has no content, an explicit message is shown in the viewer confirming that the file is empty.

​
Unified file bar

Each file viewer has a unified file bar at the top of the panel. It combines the file path and available actions — such as downloading or opening the file — into a single row, so you can identify the file and take action without scrolling.

​
Context vs Knowledge

Context injects the full file into every agent run. Knowledge uses retrieval-augmented generation (RAG) to search your files and return only the most relevant chunks. Use Context for small, essential reference files. Use Knowledge for large or frequently updated document sets.
For more on Knowledge, see Create knowledge.

​
Best practices

  • Keep files small. Context files are injected in full on every run and count toward the model’s context window limit. Large files can reduce the space available for conversation and tool output.
  • Prefer structured formats. Markdown, CSV, and code files have dedicated viewers and are generally easier for models to parse than binary formats.
  • Avoid Excel files. Excel files cannot be previewed or read directly by the model. Convert to CSV before adding to Context.
  • Review what the agent sees. Use the file viewers to confirm the file content looks correct before running the agent. YAML frontmatter in Markdown files is visible in the Properties panel.
  • Switch to Knowledge for large datasets. If your file is too large to fit in context, or if only a portion of it is relevant to any given task, Knowledge with RAG search is more appropriate.