AI API Hub: Integration, Gateways, Cost, and Risk
An AI API hub covering integration and platform selection for OpenAI, Claude, Gemini, and DeepSeek, self-hosted gateways such as One API, LiteLLM, and New API, plus caching, relay risks, and observability.
- 11 articles
- 6 wiki entries
- 7 gateway pages
Getting the first call to work takes ten minutes. Everything after that is the real work: unifying calls across models, managing keys, controlling cost, and finding the cause when something breaks. This hub collects our integration, gateway, cost, and risk content in the order you meet it.
Who is this hub for?
On this page
Start here: concepts and integration
API keys, compatible endpoints, gateways, and routing.
- Wiki
API
Wiki entry: what an API is
- Wiki
API Key
What a key does and how to keep it safe
- Article
What Is the OpenAI-Compatible API? Why Do So Many Models Support It?
From DeepSeek and Tongyi to vLLM and Ollama, nearly everyone offers an 'OpenAI-compatible' interface. This isn't an off…
- Wiki
AI Gateway
Where an AI gateway sits in the stack
- Wiki
Model Routing
Basic strategies for routing across models
Platform selection: OpenAI, Claude, Gemini, DeepSeek
How the official platforms differ, and where to start for each model.
- Article
The Ultimate 2026 AI API Platform Review: OpenAI vs. Anthropic vs. Google Gemini vs. DeepSeek vs. Kimi
A practical comparison of OpenAI, Anthropic, Google Gemini, DeepSeek, and Kimi across model quality, pricing, reliabili…
- Gateway
OpenAI API gateways
Gateways that serve OpenAI models or accept the OpenAI-compatible format. Check the base_url, exact model names, functi…
- Gateway
Claude API gateways
Gateways that carry Claude, for developers who need Claude's text, coding, long-context, and tool-calling abilities thr…
- Gateway
Gemini API gateways
Gateways that carry Gemini. Check multimodal input, file uploads, and whether the platform exposes the OpenAI-compatibl…
- Gateway
DeepSeek API gateways
Gateways and Chinese model platforms that serve DeepSeek — a fit for Chinese-language apps, coding assistants, and low-…
Gateways and self-hosting: One API, LiteLLM, New API
Unify calls across providers and pull keys, limits, logs, and billing into one layer.
- Article
What Is an AI API Gateway—and Why Shouldn't Enterprise Apps Call Model APIs Directly?
When a third team starts requesting its own OpenAI key, the company probably needs an AI gateway. This guide explains w…
- Article
Designing an AI API Gateway from Scratch: Keys, Rate Limiting, Logging, Billing, and Model Routing
Without relying on off-the-shelf solutions, build your own AI gateway: how to normalize requests, consolidate key manag…
- Article
How to Build Your Own AI API Gateway: Managing AI APIs with New API, One API, and LiteLLM
Want to centrally manage multiple AI models, API keys, usage metrics, and costs? This article clarifies the architectur…
- Gateway
One API
An open-source project for managing and distributing OpenAI-style interfaces; commonly used to self-host an API relay w…
- Gateway
LiteLLM Proxy
An open-source LLM gateway/proxy layer that unifies many providers into an OpenAI-style interface; ideal for teams that…
- Gateway
New API
An open-source multi-channel gateway evolved from the One API ecosystem, with more model channels, billing features, an…
Cost and performance
Caching, model tiering, and pricing structure — measure before you optimize.
- Article
What Is Prompt Caching? How to Use It to Reduce AI API Costs
Repeatedly sending identical system prompts and knowledge base prefixes is a hidden driver of inflated AI bills. Prompt…
- Wiki
Prompt Caching
Wiki entry: prompt caching
- Article
After the Foundation-Model Price War: Where Will Cheap APIs Take AI?
Lower prices are not the destination. They will shift competition away from inference costs and toward depth of use cas…
- Section
API cost calculator
Estimate monthly spend by usage and model tier.
Risk, compliance, and observability
Relay pitfalls, where your data goes, and how to see what happens in production.
- Article
Avoiding Pitfalls in AI Proxy Services: The Truth Behind $1 Million Tokens, Account Bans, and Model Degradation
Why have AI proxy services suddenly become so popular? What risks lie behind low-cost APIs, unified keys, and aggregate…
- Wiki
API Relay
Wiki entry: what an API relay is
- Article
AI Data Security: The More Popular the Tool, the More You Need to Know Where Your Data Goes
Before giving materials to an AI, understand data classification, retention, use in training, and access permissions.
- Article
The Ultimate 2026 LLM Observability Comparison: Langfuse vs. PromptLayer vs. Helicone vs. Arize Phoenix
A production-focused comparison of Langfuse, PromptLayer, Helicone, and Arize Phoenix across tracing, evaluation, self-…
- Article
Reimagining Enterprise API Strategy: In the Age of Agents, Systems Must Be Callable
For agents to truly get things done, enterprise systems must provide secure, auditable interfaces with appropriate gran…
Directories and lookups
The gateway directory, side-by-side comparison, and availability records.
- Section
AI gateway directory
Filter relay and gateway services by model, payment, and region.
- Section
Gateway comparison
Put candidate services side by side on models, pricing, and capability.
- Section
Pre-integration checks
The connectivity and billing checklist to run before launch.
- Section
Availability records
Access status and update dates we track.
FAQ
- Q: Should I call the official API directly or go through a relay?
- A: If you can open an official account and handle overseas billing, call the provider directly: the shortest path and the clearest accountability. Relays solve payment and access problems at the cost of an intermediary you do not control — balance, rate limits, and data retention are all on their side. Our relay pitfalls guide lists what to check.
- Q: When is a self-hosted AI gateway worth it?
- A: When any two of these appear: multiple providers to call through one interface, per-team keys and quotas, or a need for complete call logs and billing records. Our gateway design article covers the pieces to build.
- Q: Does an OpenAI-compatible endpoint mean models are interchangeable?
- A: Only at the request-format layer. Function calling, multimodal input, streaming details, context length, and rate limits still differ, so any switch needs revalidation. Our article on compatible endpoints explains where they diverge.
- Q: API spend is climbing — what should I fix first?
- A: Look at the input side first: long system prompts and repeated context usually dominate, so prompt caching and context trimming pay off fastest. Then tier your models and push simple tasks to cheaper ones. Our prompt caching guide shows how.