OLD
SaaS APIs
Predictable, Per-Call Cost
API calls have a fixed, amortizable compute cost. The size of the JSON payload has a negligible impact on the price.
Billed by the Request
The billable event is the API call itself.
SAP for AI Agents
StringCost: Think SAP for AI Agents.
Our LLM proxy simplifies AI billing by creating a double-entry ledger for every model call. Send your prompt, chosen model, and Customer ID; we handle routing to the provider, returning the response, and attributing tokens by model and type, all in one request.
Revenue-firstAudit-readyAgent-native
Andreessen Horowitz just declared:
“AI is driving a shift towards outcome-based pricing. Software is becoming labor.”
But where is the infrastructure for outcome-based billing & accounting?
This not a payments problem, but an accounting problem that leads to catastrophic failure. The future is outcome based...but the tools to build it don't exist.
Until now.
The Root Cause: A New Cost Paradigm
Why is monetizing AI agents so difficult? Because the financial models that powered the last decade of SaaS are fundamentally broken in the world of Generative AI.
VS
NEW
Agentic APIs
Variable, Computational Cost
An agent's cost is directly tied to the “work” it performs. It's a metered, computational resource, not a fixed endpoint.
Billed by the Computation
The billable events are the input tokens, output tokens, the number of thoughts, and every downstream tool call the agent makes.
This creates a massive accounting problem that traditional billing systems cannot solve. Trying to price a variable, multi-step agentic workflow like a simple API call forces you to either guess—and consistently underbill or overcharge your customers—or build a complex, brittle accounting system from scratch.
The Double-Entry Ledger for AI Agent Operations
Simple token counters and API proxies are failing. They see the token count and total bill from your model provider, but they can't tell you the profitability of a single agent run.
Deep Runtime Inspection
StringCost isn't a library that invades your code; it's a runtime-aware control plane that intelligently inspects your agent's traffic. We integrate at the network level using a secure, signed-URL architecture. Your code calls our Gateway run this request through an asynchronous deep prompt inspection engine, using a meta-classifier to understand the business intent of the action without adding a single millisecond of latency to your user's request.
+
Every Action is a Ledger Entry
Automatic instrumentation is the core of this ledger. Our Local Sidecar and asynchronous classifier automatically record every discrete computational action—whether it's an LLM call or an external tool—as a distinct line item with two sides:
- Cost Side (Debits): The tokens you paid your provider, the price of the tool call, and the compute time.
- Revenue Side (Credits): The billable event for your customer based on your pricing model for that specific action.
The result is a perfect, auditable, real-time P&L for every single agent run. You can finally answer critical business questions: “Which agents are most profitable?”, “Which tools are driving the most cost?”, and “What is the true margin on our AI features?”
Your Single Endpoint for the Entire AI Ecosystem
Why waste months integrating different providers? We've done the work so you can focus on building.
1
250+ Providers, 1 API
Access OpenAI, Anthropic, Google Gemini, Cohere, Groq, and hundreds more through a single, unified, OpenAI-compatible endpoint.
2
Write Once, Run Anywhere
Write your code in the standard OpenAI format. We build on top of production-grade open-source tooling that automatically handles the complex prompt and response transformations for you.
3
Future-Proof Your Stack
A new, cheaper model just dropped? Swap to it by changing a single parameter in your config. No refactoring, no new SDKs, no vendor lock-in.
Double-Entry Accounting
A True Ledger, Not Just a Token Counter
Your AI bill is a black box. StringCost gives you X-ray vision.
While other proxies just count total tokens, we provide a double-entry ledger for every agent run. Our asynchronous background worker inspects every prompt to give you a true P&L statement for your AI.
⚡
Zero-Latency Accounting
Our Event Collector logs the raw event instantly and returns the response to your user with zero delay.
⚙️
Asynchronous Classification
A background Worker (polling every 200ms) calls a meta-classifier to tag every request with an action_type (e.g., synthesis, tool_selection, evaluation).
💡
Know Your “Why”
Finally, you can answer critical business questions. What's the P&L of your Tree-of-Thought agent? Are “evaluation” steps costing more than “synthesis” steps? StringCost gives you the answers.
🔒Security First
Unbreakable Security: No More Static API Keys
Stop embedding sk-xxx keys in your agents, clients, or servers. Our architecture is built on a dynamic, signed-URL model that makes key leakage impossible.
1
Request a One-Time URL
Your application asks the Control Plane for permission to run a call.
2
Get a Signed URL
The Control Plane returns a short-lived, single-use signed URL that contains the encrypted credentials and user context.
3
Call the Gateway
Your agent uses this temporary URL to call the Gateway. Our system validates the signature, checks for replay attacks, and proxies the call.
The Enterprise Feature You've Been Waiting For: “Bring Your Own Key” (BYOK)
Stop subsidizing your customers' AI usage. StringCost is built for B2B, allowing your users to provide their own provider keys.
Client-Provided Keys
Let your customers enter their own OpenAI, Gemini, or Anthropic keys.
Secure & Temporary Storage
We encrypt their key at rest using pgcrypto and set a configurable TTL (e.g., 1 hour).
Automatic Cleanup
Our pg_cron job automatically and permanently deletes expired keys. You get all the benefits of BYOK without the risk or liability.
Built for Enterprise Scale & Data Control
StringCost is not a simple SaaS tool; it's a production-grade stack designed for serious enterprise deployment. You get the control of an on-premise solution with the flexibility of the cloud.
Your Cloud, Your Data
Run the high-performance Gateway and Control Plane in your own Kubernetes cluster. This ensures your sensitive prompts, keys, and AI responses never leave your network, giving you maximum security and compliance.
Flexible Hybrid Model
Use our powerful, managed Classifier service in our cloud, or deploy the entire stack—including the Worker—within your own VPC. The choice is yours.
Kubernetes-Native
Our system is a production-grade, K8s-native application, packaged with Helm for easy, repeatable deployment to any certified Kubernetes cluster, whether it's GKE, EKS, AKS, or self-hosted.
Idempotent & Safe Deployments
We ensure your infrastructure is always in a reliable state. Database migrations are built to run automatically as Kubernetes Jobs before any service starts, guaranteeing your deployments are safe, idempotent, and roll back cleanly.
Get the Control You Need and the Scalability You Deserve.
Stop building brittle, insecure, and non-monetizable AI apps. Start building on a true enterprise-grade control plane.