[Workflow memory]
A browser workflow usually needs more than a sequence of clicks. It needs operating rules, reference data, client-specific policies, and document facts. Ramain attaches those materials to the workflow datastore and lets blocks retrieve relevant context during execution.
Documents become retrievable context
Ramain can extract usable context from PDFs, text files, markdown, JSON, HTML, CSV, XLS, and XLSX files. Tables can be understood as structured rows and columns rather than treated as plain blobs of text.
Long documents are chunked with overlap and ranked by query terms, while small documents can be returned whole. This gives the agent targeted facts without pushing every uploaded file into every prompt.
Context is scoped to the workflow
In the editor, a workflow datastore card shows uploaded documents, document type, size, and viewer state. Individual blocks can scope their data retrieval capability to specific documents.
This is important for multi-client operations: the right carrier manual, rate sheet, policy, or client instruction should travel with the workflow that needs it.
How agents use it
A Datastore Agent preset retrieves relevant facts from uploaded workflow documents, applies those facts to the browser task, and includes a summary of what was used.
That turns static knowledge into runtime context without asking a human to re-explain the rules on every run.
Knowledge base context lets teams package operating rules with the workflow, so agents can apply the right instructions without repeated human prompting.
