What it is
A set of automation workflows that put the private LLM to work on everyday tasks, built in a visual workflow tool so that colleagues can read, copy and adapt them without writing code.
Each workflow is small and does one job; together they show the pattern for turning any repetitive task into an AI-assisted one.
What it does
- Document extraction: pull structured fields out of PDFs and scanned documents, one file or many at once.
- RAG ingest and ask: load documents into the vector store, then answer questions over them with sources.
- Chat agents over live systems: ask the ERP a question through a Model Context Protocol connector, or query integration audit data through a SQL tool.
- Writing help: an email polisher that keeps the author's meaning and fixes the tone.
How it works
- A trigger (a chat message, an upload or a schedule) starts the workflow.
- The workflow calls the private LLM through the same gateway every other tool uses, with a key and budget of its own.
- Tools are exposed to the model as explicit steps, so it can look things up but not act outside the workflow.
- Results go back to the chat, a file or the next workflow.
Design choices
- Visual and copyable: a workflow is something a colleague can open, understand and change.
- Same gateway, same guardrails: workflows get no special access to models or data.
- Start as proofs of concept, promote the ones that earn their keep.