Filer Usecase
Stop Starting From a Blank Page
A Filer folder agent searches your own past work for similar cases and drafts a first version grounded in what you've already written — automatically, the moment you start a new one.
- Chat
What you get
At any organization with access to a Local AI, Frontier AI (API key), or local LLM server, the same kind of report or proposal gets written over and over. Something similar almost always exists somewhere in your files — but finding it, rereading it, and pulling the relevant parts into a new draft usually takes as long as just starting from scratch.
Point a Filer folder at where that past work lives. The next time you need something similar, just ask — the agent searches your own history by meaning (not filename or keyword), finds the closest matches, and writes a new draft grounded in them, with citations back to the files it used.
How to do it
Screenshots below are from an actual run against a live Filer instance, using two synthetic sample proposals in place of real past documents.
What it needs
- A Filer folder pointed at where your past documents live
- Those documents already indexed by that folder's agent
- An inference endpoint — Local AI, Frontier AI (API key), or a local LLM server
Point the agent at your past work
Two earlier vendor-onboarding proposals sit in the folder. Asking the agent to memorize them indexes both — no separate archiving step, and the index stays scoped to this folder. Try asking something like:
Prompt
Memorize both vendor onboarding proposal files in this folder so I can search similar past proposals later.
Ask for the next one
A request for a proposal covering a new vendor triggers a semantic search over the indexed history — found by meaning, not keyword. The agent calls
search_knowledge, pulls back the closest past cases, and starts drafting from what it finds. For example:Prompt
I need to draft a proposal for <new vendor>. Search my past proposals for similar cases and draft it grounded in what you find.
Get a draft grounded in what it found
The agent carries over the same structure both past proposals shared — overview, scope, timeline — and adapts the specifics to the new vendor. You still review and finish it — the agent gives you a grounded starting point, not a black box.
What the agent actually wrote "Bright Path Textiles will undergo a comprehensive onboarding process covering purchase-order integration, quality checks, and a 90-day delivery cadence... Timeline — Discovery (1 week): needs assessment, stakeholder meetings, current system audit. Integration (3 weeks): system configuration, data mapping, process definition. Pilot (2 weeks): limited production run, quality verification, optimization review."
Verified
This isn't a mockup — the screenshots above are from a live run. The retrieval that powers it — indexing documents and searching them by meaning — is the same pipeline Filer verifies against a live instance on every release. Drafting from the retrieved content is regular language-model reasoning, not something unique to this workflow.
Where else this applies
Any recurring document type where a past example is a better starting point than a blank page fits the same memorize-then-search-and-draft shape — vendor onboarding is one instance, not the whole scope:
- Client proposals — a new pitch grounded in the closest-matching past win, not started from a generic template.
- Incident postmortems — a new writeup drafted against the structure of the closest past incident, not reinvented each time.
- Support replies — a first draft grounded in how a similar ticket was actually resolved before, not a generic canned response.
Other ways in
Every other trigger source in Filer — a file landing, a schedule, an inbound message — fires the same way every time. This one can't: what "similar" means depends entirely on what you're about to write, so there's nothing for a file-watch or a cron schedule to key off. Chat is the entry point here, not a fallback — just ask, whenever you're starting the next one.
- Right-click
- There's no bundled right-click action for this yet — chat is the way in, same as the walkthrough above.
- Chat
- This is the whole case, in its simplest form — no folder pointer, no memorize step spelled out:
"Find something similar I've written before and use it as a starting draft for this." - Other automation triggers
- None — see "Why this one is chat-only" above. What counts as "similar" depends on what you're about to write, so there's nothing for a file-watch or a schedule to key off.
- Other integrations
- This works over whatever's already indexed in the folder — it doesn't matter whether those past documents started life as local files, an email attachment, or a Notion export. Getting them memorized in the first place is the only step that depends on where they came from.