Use Cases and Playbooks
Agents in Assist are good at one-shot questions. Agents with a filesystem are good at work that takes days, weeks, or runs on a schedule — because they have a place to put what they learn and find it again later. These playbooks walk you through building real systems where the filesystem is the persistent memory that makes the workflow possible.
Each playbook combines the filesystem with the rest of the Assist platform — custom tools that connect to your internal systems, sandcastle apps your team can use without writing code, and subagents that run on schedules. You build all of this from inside your AI client through conversation. You do not need to log into Assist to follow any of them.
Playbooks
Build a Vendor Research Agent That Remembers Every Session
Your procurement team researches the same vendors over and over. Every conversation with an AI starts from zero, and the analyst ends up re-explaining what they already know. Build a subagent with its own filesystem that keeps a file per vendor, a decision memo for each evaluation, and a master index. Every new chat with the agent picks up exactly where the last one ended. Add a sandcastle app so the whole procurement team can browse the research without opening a chat.
Build a Monthly Reporting Agent With Templates and an Archive
Someone on your team generates the same report every month from your ERP or analytics system. The format drifts, the numbers get copy-pasted, and last month's report is buried in an email thread. Build a scheduled subagent that reads a template file, calls tools to pull this month's numbers from the source of truth, writes the completed report to a dated file, and maintains a running archive. Add a sandcastle dashboard that shows every past report with a one-click compare.
Build a Reconciliation Drift Tracker Across Two Systems
Your warehouse management system and your ERP are supposed to agree but they do not. Someone runs a manual reconciliation every month and by the time they find a discrepancy nobody remembers whether it is new or old. Build a scheduled subagent that pulls daily snapshots from both systems into per-day files, diffs them to produce a discrepancy log, and tracks which mismatches are new, recurring, or resolved. Add a sandcastle app where the ops team investigates each discrepancy and marks it done.