Freelancers are bleeding revenue, losing an estimated 15% to 20% of billable earnings every week due to unbilled micro-tasks, vague invoices, and silent scope creep. PhyInvoice AI, a new submission for the Hacktoberfest Open-Source AI Challenge, aims to plug these leaks with a local-first billing copilot. Built by developer shivamgravity, the tool combines open-weight language models with tabular forecasting to create a revenue defense engine that runs entirely on the user's laptop, eliminating the need for cloud API subscriptions that erode small business margins.

Local AI for Confidential Client Work

The core technical challenge for solo contractors is the Non-Disclosure Agreement (NDA). Pasting proprietary client tickets or internal architecture discussions into commercial LLM clouds often breaches confidentiality terms. PhyInvoice AI solves this by automatically detecting locally installed Google Gemma models via Ollama. It transforms rough developer notes, such as "debugging jwt auth," into legally defensible, professional deliverable statements like "Resolved asynchronous JWT authentication token refresh race condition on WebKit/Safari viewports." This ensures that sensitive client data and billing rates never leave the freelancer's machine, addressing a critical compliance pain point for developers working under strict NDAs.

Predicting Scope Creep with Tabular Models

Beyond text generation, the tool tackles financial leakage using Prior Labs' TabPFN-3.5, a tabular foundation model. While LLMs handle language, TabPFN excels at predicting outcomes from small numerical tables without fine-tuning. The engine extracts eight specific features per project, including budget utilization, duration variance, and off-hours ratio, to forecast budget overruns. For example, a high frequency of tasks containing scope-expansion keywords like "emergency" or "slack" triggers a warning about projected financial loss. This predictive capability allows freelancers to identify secretly unprofitable projects before they become a financial burden, moving beyond reactive billing to proactive revenue management.

Deterministic Billing and Open Source Accessibility

The architecture strictly separates AI from arithmetic to ensure accuracy. A deterministic agency billing engine enforces standard tenth-hour (6-minute) and quarter-hour (15-minute) rounding increments, calculating the true Effective Hourly Rate (EHR) for each project. The project is fully open source under the MIT License, with a live demo available at phyinvoice-ai.onrender.com. Users are encouraged to disconnect their Wi-Fi during the demo to prove the 100% local-first nature of the application, which relies on IndexedDB for storage and local models for processing. The pricing model for the hosted version is set at $19/month or $49 lifetime, targeting the market segment that finds existing SaaS tools like Harvest or Toggl too expensive for micro-business margins.

Key Takeaways

  • PhyInvoice AI uses local Google Gemma models via Ollama to generate professional invoice descriptions from rough notes, preserving client NDA compliance.
  • Prior Labs TabPFN-3.5 is integrated to predict scope creep and revenue leakage by analyzing historical session variance and off-hours work ratios.
  • The tool is 100% local-first, allowing freelancers to track time and generate invoices without an internet connection, reducing cloud dependency.
  • Open source under MIT License, it offers a cost-effective alternative to $20-$40/month SaaS billing platforms for solo contractors.

The Bottom Line

PhyInvoice AI correctly identifies that for freelancers, privacy and predictability are non-negotiable features. By keeping the entire billing pipeline local, it solves the NDA compliance issue that plagues cloud-based competitors while offering predictive insights that static time trackers simply cannot provide.