99.2% on WebVoyagerState-of-the-art browser and computer-use agents
Research/Platform · 2026-10-18 · 4 min read

Knowledge base context: giving workflows durable memory

Real work runs on rules, reference tables and client instructions, not only clicks. Ramain keeps that material with the workflow, so a digital worker applies it on every run without being told again.

Key points
  1. 01Policies, rate tables and instructions live in the workflow's datastore.
  2. 02A digital worker looks up the facts a step needs instead of reading every file every time.
  3. 03Request-specific files stay with the run that uses them, so one case never leaks into the next.

Rules travel with the workflow

A workflow that only records clicks breaks the first time a case needs judgment. The person doing the job knows which carrier manual applies, which fee schedule is current and what this client asked for last quarter. A digital worker needs the same material.

In Ramain that material lives in the workflow's datastore: documents such as PDFs, spreadsheets and text files that the workflow can consult while it runs.

Scoped to the work that needs it

Reference material belongs to the workflow that uses it, and a step can be pointed at the specific documents it should consult. That matters in multi-client operations, where the right manual, policy or instruction has to travel with the right workflow and no other.

How a digital worker uses it

When a step needs a fact, the digital worker retrieves the relevant passages rather than loading every file into every step. It applies what it found to the task in the browser, and the run is recorded, so a reviewer can see what it did.

What goes where

  1. 01Reusable reference material, such as a checklist, policy or rate table, goes in the datastore.
  2. 02Values that change on every run, such as a loan number or a customer name, are launch inputs.
  3. 03A document that belongs to one request, such as this customer's application PDF, stays with that run.

Keeping one example invoice in the reusable datastore would teach the workflow to reuse it. Keeping request files with their run avoids that.

PlatformDatastoreWorkflow design

See Ramain run your own workflows.

Bring a process with rules your team applies by hand. We will show a digital worker applying them.

Book a demo