Pylva: An Open-Source Usage Billing Platform Built for AI Agents

A Judgment from the Developer Community

Recently, an independent developer posted their open-source project Pylva on Reddit and invited public feedback. The project itself is worth attention, but what's more interesting is the reasoning behind it: SaaS will ultimately be replaced by AI Agents — founders will no longer sell software subscriptions, but instead build agents and sell the "outcomes" they produce.

After months of deep research, this developer came to hold that belief firmly. But when they actually entered the space, they found the market lacked robust solutions to help AI Agent builders scale and monetize with ease. That's how Pylva was born.

This isn't an isolated take. The paradigm shift from "Software as a Service" (SaaS) to "Service as a Software" is becoming a hot topic across Silicon Valley and the broader tech investment community. The concept was formally articulated around 2024 by Andreessen Horowitz (a16z) partners Martin Casado and Sarah Wang: traditional SaaS sells "the right to use software," while AI Agents sell "the completion of a task." This shift moves the pricing anchor from "seat count" or "feature access" to "actual business value delivered." Salesforce CEO Marc Benioff has also stated publicly on multiple occasions that AI Agents will disrupt the per-seat SaaS era. This mirrors every major computing paradigm shift in history — from mainframes to personal computing to cloud computing — each democratization and intelligence leap has been accompanied by a fundamental restructuring of business models. When AI Agents can directly complete tasks and deliver results, the traditional per-seat billing model simply doesn't hold up.

What Pylva Is: Three Core Capabilities

Pylva is positioned as a fully open-source, self-hosted AI Agent billing platform, hosted on GitHub and purpose-built for AI Agent builders. Its core functionality can be broken down into three areas:

Full-Funnel Usage Tracking

Pylva lets developers track usage across all customers. In AI Agent monetization, this capability is critical — every task execution, every API call, every token consumed maps directly to real compute costs. Without granular usage tracking, profitability is a fantasy.

Flexible Per-Customer Billing Rules

The platform supports flexible usage rules per customer. Developers can configure independent metering and quota policies for different customer tiers and payment plans, enabling fine-grained commercial operations.

Automated Invoice Generation

Pylva automatically generates invoices for customers. For indie developers and small teams looking to quickly monetize their agents, this saves enormous engineering effort that would otherwise go into building a billing system from scratch.

Why Usage-Based Billing Is Essential in the AI Agent Era

Traditional SaaS subscriptions and per-seat pricing were built on a relatively stable and predictable cost structure. The economics of AI Agents are fundamentally different.

LLM inference costs are variable and usage-based. When an agent handles a complex task, it may trigger dozens of model calls and consume millions of tokens. If builders stick to flat-rate subscription pricing, heavy users can quickly turn profitable products into money-losing ones.

It's worth noting that usage-based billing isn't a new invention of the AI era — it's a well-validated model from cloud computing. AWS's per-second EC2 billing, Twilio's per-message SMS API, and Snowflake's compute-unit pricing are all classic examples. OpenView's annual reports consistently show that SaaS companies using usage-based pricing outperform pure subscription companies in growth rate and net revenue retention (NRR). At the AI infrastructure layer, OpenAI, Anthropic, and Google all charge developers per token, which means AI Agent cost structures are inherently variable — and that pushes application-layer business models toward usage-based pricing as well.

So usage-based billing is almost an inevitable choice for AI Agent monetization. Pylva targets this pain point directly: freeing developers from the complexity of metering, quota management, and invoicing logic so they can focus on building better agents.

From this angle, Pylva is solving a problem with a clear lineage — Stripe in payments, and Lago or OpenMeter in metering and billing. Lago is positioned as an open-source alternative to Stripe Billing, focused on modeling complex billing logic, and raised a $6M seed round in 2022. OpenMeter focuses on high-performance real-time event metering, built on Apache Kafka, and handles metering scenarios with millions of events per second. Both projects validate the same market thesis: billing infrastructure is complex enough and general enough to exist as a standalone product. Pylva's differentiation strategy on top of this is vertical focus on the AI Agent use case — offering native support for LLM token consumption, agent task invocations, and other AI-specific metering dimensions, rather than asking developers to adapt a general-purpose billing framework from scratch.

Open Source and Self-Hosting: Strengths and Challenges

Pylva's open-source, self-hosted approach is a choice full of meaningful trade-offs.

The core advantages: data sovereignty stays entirely in the developer's hands. A billing system naturally aggregates sensitive data — customer usage behavior, payment capacity, and business scale. Under increasingly strict data privacy regulations like GDPR and CCPA, this data requires tight control. For developers serving European markets, keeping data on-premises and out of third-party SaaS platforms is often a compliance requirement, not an option. In B2B contexts, enterprise security reviews routinely ask where billing and usage data is stored — a self-hosted solution significantly reduces friction at that stage. The sustained enterprise success of data sovereignty-focused tools like HashiCorp Vault and GitLab validates this approach. Open source also lowers the trust barrier: developers can audit the code and customize it freely.

The real challenges are equally significant: self-hosting requires developers to handle deployment, operations, and scaling on their own, which raises the initial barrier for teams that want something out-of-the-box. How to build a sustainable business model — such as the Open Core model of open-source plus a hosted cloud service — is also a long-term challenge.

A Clear-Eyed View of an Early-Stage Project

To be honest, Pylva is still at a very early stage — it's a project launched by an independent developer who is actively seeking feedback, and the author even invites interested users to reach out via DM to help.

This means:

The project's maturity, stability, and community size are yet to be proven;
Its sweeping claim that "SaaS will be replaced by AI Agents" is more of a forward-looking bet than a market-validated fact;
Compared to mature billing solutions, Pylva still needs significant work on feature completeness and ecosystem development.

Conclusion: Betting on a Future Still Taking Shape

The story of Pylva is fundamentally about a developer making a deliberate bet on future trends. It's betting on a world where the "Agent economy" gradually takes shape — a world where countless founders build all kinds of intelligent agents and charge for the outcomes they actually deliver.

Whether or not that bet ultimately pays off, the problem Pylva points to is real: as AI evolves from an assistive tool into a "digital worker" capable of independently completing tasks, how do we price its labor? Usage tracking, flexible billing, automated invoicing — these may sound like boring infrastructure, but they are the indispensable foundation on which new business models must be built.

For developers exploring AI Agent monetization paths, Pylva may be worth trying — or at the very least, worth keeping an eye on. And for the industry as a whole, the emergence of tools like this is itself a powerful signal that the "Agent economy" is moving from concept to reality.