VaultLedger AI: The 100% Local, Air-Gapped Freelance Financial & Invoice Intelligence OS (Zero Cloud Data Leakage)

Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle Submission 💸

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle

If you are a freelancer, independent software consultant, or solo creative, managing your finances is a weekly headache. You juggle PDF client invoices, messy restaurant receipts, software subscription bills, and subcontractor invoices.

To make matters worse, freelancers face two huge obstacles:

VaultLedger AI is an air-gapped financial operating system built specifically for freelancers and solopreneurs. It runs 100% locally on your laptop using open-weight AI ( Llama 3.2 3B via Ollama ).

Freelancers would gladly pay a $49–$79 one-time license (or $9/month self-hosted subscription) for unlimited, completely private, offline bookkeeping.

VaultLedger AI features a dark glassmorphic interface with an emerald security aesthetic:

Displays real-time connection status to local Ollama ( llama3.2:latest ), Gross Invoiced Revenue, Total Business Expenses, IRS Deductible Write-Offs, and Estimated Self-Employment Tax Savings (at 30%).

Search and filter across all processed transactions. One-click export generates an RFC 4180-compliant IRS Schedule C CSV mapped directly to lines 11, 13, 18, and 24b.

The project is structured with a local Node.js Express backend mediating communication with Ollama, and a lightweight Vanilla HTML/CSS/JS frontend.

// server.js - Contextual payment collection email drafter app. post ( ' /api/draft-email ', async ( req, res ) => { const { invoice, tone = ' gentle ', daysOverdue = 14, model = ' llama3.2:latest ' } = req. body; const prompt = `You are an expert freelance business advocate. Write a polite, effective, professional payment follow-up email from a freelancer to a client for an overdue invoice. Details: - Client / Company Name: ${ invoice. vendor_or_client } - Invoice #: ${ invoice. invoice_number } - Amount Due: ${ invoice. currency }${ invoice. total } - Days Overdue: ${ daysOverdue } days - Tone: ${ tone } (Options: gentle reminder, professional follow-up, firm final notice) Return ONLY valid JSON: { "subject": "Clear email subject line", "email_body": "Complete email text formatted with salutation, body paragraphs, and professional sign-off" }`; const ollamaResponse = await callOllama ( ' /api/generate ', { model: model, prompt: prompt, format: ' json ', stream: false, options: { temperature: 0.3 } }, 30000 ); const result = JSON. parse ( ollamaResponse. response ); res. json ({ success: true, email: result }); }); Enter fullscreen mode Exit fullscreen mode 🛠️ How I Built It 1.

Open-Source AI Stack Open-Weight Model: Llama 3.2 3B ( llama3.2:latest ), Meta's state-of-the-art compact multilingual model. It runs comfortably on consumer hardware with 2GB RAM footprint (quantized to Q4_K_M GGUF format). Inference Harness: Ollama (v0.40.3) running locally via HTTP on http://127.0.0.1:11434.

Ollama provides high-throughput CPU/GPU acceleration, memory-mapped weights, and structured JSON output mode ( format: 'json' ). Prompt Architecture: Strict JSON schema prompting with few-shot financial classification context and IRS Schedule C tax guidelines embedded into system instructions. 2.

Web Stack & Design System Core: Node.js and Express backend providing an air-gapped local mediator with persistent on-disk JSON storage ( data/ledger.json ). Styling: 100% Vanilla CSS3 using custom CSS design tokens, glassmorphism, responsive grid layouts, and typography powered by Google Fonts ( Outfit and Plus Jakarta Sans ). Zero Third-Party Telemetry: No Google Analytics, no tracking pixels, no external CDNs that could leak metadata. 🌍 Why Does Open Innovation Matter?

Why does open innovation matter for what I built? What did an open-source approach make possible that a closed API wouldn't?

When a freelancer works for an enterprise client or healthcare startup, they routinely sign agreements forbidding the upload of project details to public cloud AI services. Because VaultLedger AI uses an open-weight model running locally, not a single character leaves the computer. It gives freelancers the superpowers of AI without putting their contracts in jeopardy.

Closed APIs charge per token. A busy freelancer or small agency auditing 200 invoices, receipts, and line items a month would burn through $30–$50/month in API tokens and SaaS subscription fees ($360–$600/year). With open-source Llama 3.2 running on the laptop, the marginal cost of running 1 or 10,000 extra extractions is literally $0.00.

Closed APIs regularly deprecate model checkpoints, change pricing tiers, alter moderation filters, or suffer outages. With open-source AI, the model belongs to the developer and user forever. If you want to swap Llama 3.2 for Mistral, Gemma 2, or a domain-specialized tax fine-tune, you can change the model parameter in one second without rewriting a line of business logic.

Freelancers work on planes, in train cars, and at remote cabins without stable internet. Closed cloud APIs fail instantly the moment the Wi-Fi drops. Because VaultLedger AI runs locally, you can categorize a year's worth of business receipts at 35,000 feet with Wi-Fi disabled.

This entire project was planned, designed, coded, and tested in collaboration with Antigravity (Google DeepMind's advanced agentic coding assistant).