Knowledge Library RAG: Ground Your Desktop Marketing Agent in Your Own PDFs and Docs
Generic AI outreach sounds like everyone else's pitch. The fix is not a longer prompt — it is your product PDFs, pricing guides, case studies, and docs sitting in a local knowledge base the agent can search before it writes.
aiFetchly's Knowledge Library is local-first document RAG for desktop marketing and outreach. Upload files or import public pages; aiFetchly chunks them, embeds them, and retrieves the right passages when AI Email Writer or AI Chat runs with RAG context on. Docs: Knowledge Library. Website import: Import Website. Features: aifetchly.com/features. Intro: Getting Started. Source: GitHub.
What Knowledge Library is
RAG (Retrieval-Augmented Generation) means:
- Ingest your documents and split them into chunks
- Embed each chunk so semantic search works (not just keywords)
- Retrieve the most relevant passages for the current prompt
- Ground the AI reply in those passages instead of inventing product details
Traditional assistants invent generic copy. With RAG, aiFetchly references your brochure, your refund policy, your competitor comparison — so outreach stays accurate and on-brand without pasting the whole PDF into every chat.
Supported formats
| Format | Extensions | Best for |
|---|---|---|
.pdf |
Brochures, whitepapers, docs | |
| Word | .doc, .docx |
Proposals, product briefs |
| Text | .txt |
Notes, plain exports |
| Markdown | .md |
READMEs, technical docs |
| HTML | .html, .htm |
Saved articles, web pages |
Processing is automatic after upload: save → chunk → embed → status moves Pending → Processing → Completed. Small files often finish in under a minute; larger PDFs take longer.
Upload files and import websites into the same pipeline
Files: open Knowledge in the left nav, then drag-and-drop or use the file browser. Multi-select works. Each document shows name, title, status, type, size, and actions (view, download, delete, re-embed, logs).
Websites: you can also import public URLs. aiFetchly fetches the page, converts it to markdown, and indexes it through the same RAG pipeline — so a pricing page and a PDF brochure are searchable together. Options include a single page, a list of URLs, or a bounded same-origin crawl. Details and limits: Website import.
Search and filter the library by name, status, or type. Bulk delete when you clean out outdated material. Re-embed after you change embedding models so older docs catch up.
Local free embedding vs remote — and privacy
Embeddings turn text into vectors for semantic search. In Knowledge Library → Settings you pick the model:
| Type | Example | Notes |
|---|---|---|
| Local (free) | Xenova/all-MiniLM-L6-v2 (free) |
Runs on your device (384-d). Private, free, offline once cached. |
| Remote | Server-provided models | Generated on aiFetchly's AI server (plan-dependent). |
With the local model selected, document text is not sent to the remote embedding endpoint — Transformers.js builds vectors on your CPU. First use downloads model weights from Hugging Face and caches them; later runs are fast.
If a remote model fails after retries, aiFetchly can fall back to the local free model for that document (quota/billing errors do not silently fall back). Documents stay in a local SQLite store; original files remain on your machine. Your library is not used to train public models.
That matters for marketing teams: product roadmaps, confidential pricing, and client-specific notes can ground outreach without shipping the whole corpus to a third-party agent marketplace.
Powers AI Email Writer and AI Chat
Knowledge Library is not a separate toy — it feeds the features you already use:
- AI Email Writer — enable RAG context; the writer searches the library and personalizes emails with product specifics from your docs.
- AI Chat / Marketing Assistant — toggle the RAG context control (📖), then ask "What are our key features?" or "Draft outreach for Product X" and answers draw on your uploads.
Example: upload a product catalog PDF, turn RAG on, and generated emails mention the SKUs and benefits that match the prospect — without you retyping the catalog into the prompt.
Practical tips
Upload detailed product docs, value props, case studies, and industry language. Avoid stale generics, irrelevant dumps, and huge files over ~10 MB (split them). Use clear names (Pricing_Guide_v2.docx). Keep the library healthy: remove outdated docs, re-upload when facts change, and check Error statuses in logs.
When a document is not showing up in AI output, confirm status is Completed, RAG is enabled in the writer/chat, and the query is specific enough to retrieve the right chunks.
Why this angle for desktop marketing agents
A desktop agent that can research, draft, and call tools is only as trustworthy as its context. Knowledge Library RAG puts your PDFs and docs under the agent — locally when you choose the free MiniLM path — so AI Email Writer and AI Chat stay grounded without turning your marketing library into someone else's training set.
Try it: aiFetchly · Features · Knowledge Library docs · Website import · GitHub