Knowledge Base (uploads, search, governance)
Upload your own rules and documents to teach your AI how your firm works.
🚀 Why This Matters
A generic AI might know about VAT in theory. But your firm has its own way of handling invoices, clients, and exceptions. The Knowledge Base is where you give your AI that context — so instead of vague advice, it gives you your rules, your policies, your way of working.
Impact you’ll notice right away:
Accuracy jumps from ~60% → 95%+
Fewer re-checks (errors cut by 70–80%)
Faster answers that reflect your internal policies
Audit confidence — AI cites your uploaded rules, not generic guesses
🔎 Pain point solved: No more “That’s not how we do it.” Every answer aligns with your procedures.
📤 How to Add Content to Your Knowledge Base
Step-by-Step Upload Process
Navigate to Knowledge Page: Click "Knowledge" in your main navigation
Click "Add Content": Use the button to open the upload modal
Choose Your Method:
📝 Text Content: Type or paste rules, policies, procedures directly
📁 File Upload: Drag & drop or select PDFs, Word docs, Excel files
Add Details:
Title: Clear, searchable name (e.g., "Client ABC VAT Policy 2024")
Tags: Categories for easy filtering (e.g., "VAT", "Client-ABC", "Policy")
Save: Content is processed and ready for AI queries within minutes
You can then view the content at any time by clicking on the view button when hovering over a piece of content.
Supported File Types
✅ PDFs: Tax circulars, policy documents, client agreements ✅ Word Documents: Internal procedures, templates, checklists ✅ Excel Files: Chart of accounts, vendor lists, rate tables ✅ Text Files: Quick notes, email chains, meeting minutes
Content Quality Tips
🎯 Be Specific: "VAT exemption for software exports to EU" > "VAT rules" 🎯 Use Examples: Include real scenarios with correct outcomes 🎯 Stay Current: Date your documents clearly (e.g., "2024_VAT_Rates.pdf") 🎯 Clear Language: Avoid jargon — write like you're training a new employee
🔒 Why Knowledge Can't Be Deleted from AI Models
The Technical Reality
Once your AI model learns from your content, that knowledge becomes part of its "neural pathways" — like how you can't "unlearn" riding a bike. Here's why:
How AI Learning Works:
Content gets processed into mathematical patterns (embeddings)
These patterns merge with millions of existing connections
The model's "understanding" becomes a blend of everything it has learned
Removing one piece would require retraining the entire model from scratch
Business Implications
Why This Actually Protects You: ✅ Consistency: Your AI won't "forget" important rules during updates ✅ Reliability: Once trained on your procedures, they're permanently embedded ✅ Audit Trail: All learning is logged and traceable for compliance
What You Can Control:
✅ Override: Upload newer versions that take priority
✅ Archive: Mark content as outdated (AI will prefer newer rules)
✅ Context: Add notes explaining when/why rules changed
Best Practices for Updates
Instead of "deleting" outdated content:
Version Control: Upload "PolicyName_v2.pdf" alongside the original
Clear Dating: Use consistent date formats in titles and content
Archive Tags: Tag old content as "Archive_2023" so it's deprioritized
Replacement Notes: Include "Replaces PolicyName_v1" in new documents
The Upside
This permanence is actually a feature:
📚 Institutional Memory: Your AI retains historical context
🔍 Reference Library: Can explain why rules changed over time
🎯 Learning Continuity: Builds on past knowledge rather than starting over
💡 Pro Tip: Think of it like training a human employee — you can teach them new ways, but they remember the old ways too. The key is making clear which approach to use when.
📥 What You Can Upload
Think of the Knowledge Base as your AI’s library. Anything you upload becomes a reference it can pull from in seconds.
Examples:
Tax & Compliance: AADE circulars, myDATA specs, VAT guidance
Internal Policies: How your firm classifies expenses, approval workflows, error-handling steps
Client Rules: Per-client preferences (e.g., “Vendor X invoices always go to account 601”)
Reference Materials: Vendor lists, chart of accounts, exchange rate policies
Training Docs: Perfect invoices, common mistakes with corrections, templates
💡 Pro Tip: Start with 10–20 “gold standard” documents (your best examples) — they teach the AI far more than 100 messy ones.
🗂 How to Organise Your Knowledge
The more organised your uploads, the sharper the AI’s answers.
By Document Type:
Sales invoices, purchase invoices, credit notes, receipts
Each type tagged so AI knows how to apply rules
By Client:
Client A → special VAT treatment
Client B → uses custom chart of accounts
Client C → multi-currency handling
By Rule Category:
Tax rules (VAT, exemptions, classifications)
Internal policies (approvals, exceptions)
Quality control (review checklists, error patterns)
By Date/Version:
“VAT_Rates_2023.pdf” → Archived
“VAT_Rates_2024.pdf” → Current
🔎 Example: Upload your “Expense Classification Guide” → AI will use your exact categories when suggesting fixes, instead of generic defaults.
🔍 Smart Search in Practice
Once you’ve uploaded your rules, you can ask everyday accounting questions in plain English or Greek.
Example questions:
“What VAT exemption applies to software sold to an EU business?”
“Which category should I use for restaurant receipts for Client X?”
“Show me our approval workflow for expenses over €5,000.”
The AI doesn’t just guess — it looks in your Knowledge Base, finds the relevant rule, and cites it back to you.
✅ Governance Made Simple
To keep your AI sharp, treat your Knowledge Base like a living library:
Update regularly: Swap out outdated guidance (don’t keep two versions).
Tag smartly: Use clear labels by client, doc type, or rule.
Archive history: Keep old rules for reference, but don’t let them clutter answers.
Review quarterly: Make sure everything reflects how you actually work today.
🛡️ Privacy & Control
Your uploads are private to your firm — never shared with anyone else.
The AI uses them to answer your questions but does not train global models.
You can delete or update documents anytime.
📈 What You’ll Gain
Day 1: AI starts applying your firm’s policies instead of generic ones.
Week 1: It learns your recurring patterns (e.g., vendor X always missing VAT).
Month 1: It feels like a junior accountant trained in your exact way of working.
💡 Pro Tips
Start with the rules you repeat most often (VAT treatments, expense categories).
Upload per-client exceptions so the AI stops treating everyone the same.
Use templates and checklists — AI will apply them consistently.
Give your documents clear names (“ClientB_VAT_Policy.pdf” beats “Document1.pdf”).
👉 In short: The Knowledge Base is how you turn your AI into your assistant — one that knows Greek tax law and your firm’s way of working.
Last updated