About Services Tools Blog Contact us

Agent Document Search

Indexes a named project folder so a coding assistant can search it for the most relevant passages without the notes being pasted into the chat.

  • Semantic Search & RAG
  • Cross-sector
  • Tool
  • Product
  • Local/private
  • 2025

The problem

Project notes are too large to paste into an assistant, and keyword search misses paraphrases. So every new chat starts as if the folder had never been read.

What it does

Documents for a named project sit in a folder and are indexed by meaning using Ollama. An assistant can list the projects and search one, through a local connection or a small PHP example. The reply is a ranked list of passages with scores, for the assistant to quote in its own work.

Outcomes

  • An assistant can search a project folder without the notes being pasted in.
  • Hits are passages with scores, ready to quote.
  • Projects stay separate, so one query searches one folder.

Simplified mockup

Interactive preview with sample data. Nothing is sent to a server.

Projects the agent can see

Listed
2
names returned to the agent
Ready to search
1
fen-catalogue has a store
Listed, not searched
1
holt-handover has no store
Project list
ProjectFolderNotesStore
fen-cataloguenotes/fen14indexed
holt-handovernotes/holt6no store
Last agent call
list_projects → fen-catalogue → holt-handover query is scoped to one name holt-handover is visible, and is not searched

Query one project

Question passages, not an essay
Retrieval
Choose Search to retrieve passages

Ingest a project folder

Folder
Click to load the fen-catalogue folder

One project name, one folder. Other projects stay in their own stores.

Ingest
Load the folder to start

How it works

  1. A project name points at a folder of documents.
  2. The files are indexed by meaning with Ollama and saved.
  3. An assistant lists the projects it can see.
  4. A search returns the best-matching passages for that project only.
  5. The same search is available to a local connection and to a PHP example.

Pipeline

flowchart LR
  Folder[Project folder] --> Ingest[Embed documents]
  Ingest --> Store[Vector store]
  Agent[Agent query] --> List[List projects]
  List --> Store
  Store --> Hits[Ranked passages]