Knowledge Base
The workspace library of files your agents read at run time.
The Knowledge Base is your workspace's shared library of files. You upload a document once — a spec, a design reference, an API doc, a screenshot — and it stays in the workspace for every agent and every teammate to draw on. Context lives in the system, not in someone's head.
Nothing in the Knowledge Base is sent anywhere on its own. Files only reach an agent when you attach them to a run, and they're staged into the agent's working directory on your own machine right before it starts.
What you can store
The same limits apply to Knowledge Base files and to the one-off files you attach to a single spec or issue.
-
Formats — PDF, Markdown, plain text, HTML, and images (JPEG, PNG, WebP, GIF, SVG).
-
File size — up to 10 MB per file.
-
Workspace capacity — up to 1 GB of files across the whole workspace.
How agents use it
When you start a Spec, you compose the request and then pick which Knowledge Base files the agent should read — a searchable list scoped to your workspace. The files you choose are staged as required reading: the agent reads every one before it works. The same files are re-staged when the build runs, so anything the spec points at is still there at build time. See the Spec agent for where this fits in a run.
One-off files (not the Knowledge Base)
You don't have to add a file to the library to use it once. Every spec and issue takes its own attachments — up to 20 files, 10 MB each — that live with that one item and are removed with it. This is how you hand the Issue Diagnosis agent a screenshot or a log to show a problem, or attach a file to confirm a fix when you keep working on a build. Issues use attachments only; they don't draw on the Knowledge Base.
How a run sees your files
The Knowledge Base files you picked, plus any one-off attachments. The agent reads every file here.
Your planning docs — brief, research, plan, brand and technical guides — added automatically when they exist. The agent consults them as needed.
Before a run starts, Fragua stages your files into the agent's working directory on your machine: required reading in knowledge/, optional context in references/.
Both surfaces that read your files start from a feature or a bug.