Upload the files
The refund policy, the tone guide, the product sheet. Up to 5 MB each.
Product · Build
Ground answers in your own files.
Attach knowledge files to agents and projects. Their text joins every chat, so answers draw on your material.
The agent stopped guessing your refund policy the week the actual document rode into every chat.
- PDF, Word, Markdown, CSV, JSON, and text
- Text extracted on upload, applied automatically
- Up to 10 files per agent, more per project
10 files per agentof your own material read into every conversation, with no per-chat setup
How it works
We extract the text
PDF and Word files are converted to plain text at upload, stored with the file.
Answers use it
Every chat with that agent or project includes the material. No setup per conversation.
01
The model answers from your documents.
A support agent that has read your actual refund policy gives a different answer than one guessing from the internet. Files make the difference.
- Knowledge applied to every chat automatically
- Works with agents and with projects
- Files stay private to your workspace
02
Simple by design.
This is direct file grounding: the leading text of your documents rides along with the conversation. No pipeline to configure, nothing to re-index by hand.
- Deterministic: the same text every turn
- Upload once, replace whenever the doc changes
- File status visible in the agent editor
Questions, answered
PDF, Word, Markdown, CSV, JSON, and text, up to 5 MB each. We extract the text at upload and store it with the file.
No. The leading text of your documents rides along with the conversation, the same text every turn. Replace the file when the document changes.
Size it with a free tool
Plan a retrieval chunk budget
Split a context window between system prompt, files, and output.
Check what fits in a context window
Test whether a document fits the model you plan to use.
The $100 AI stack before your first operator
What a small team can run before anyone is hired to run it.