Comparison
Supercompat sits in a specific slot in the multi-provider LLM space. Here's how it compares to the closest alternatives.
At a glance
Supercompat vs. Vercel AI SDK
Vercel AI SDK gives you its own abstraction —
generateText, streamText, generateObject — over every provider. You write against ai, not openai or @anthropic-ai/sdk.Supercompat gives you the real
OpenAI SDK (or Anthropic SDK). You write against the provider's own SDK and get the provider's full native feature surface — not a lowest-common-denominator abstraction. Threads, runs, Responses conversations, server-side web search, file search, code interpreter, computer use, typed event streams, and every SDK utility function are all in scope — because you're calling the real SDK.Supercompat is the better pick when you need the full native depth of a provider's API, regardless of which provider is behind it. Vercel AI SDK optimizes for a unified app-level abstraction; Supercompat optimizes for preserving the provider SDK experience while still letting you swap backends.
Supercompat vs. LangChain
LangChain is a higher-level framework: chains, agents, retrievers, memory modules. Its model abstraction is one piece of a much larger surface.
Supercompat is narrower and lower-level. It doesn't define chains or agents; it gives you a real LLM SDK pointed at any provider, with storage and run adapters for the Assistants and Responses surfaces.
Pick LangChain when you want the whole orchestration framework.
Pick Supercompat when you want the SDK plus a thin, typed compatibility layer — and you want to compose your own orchestration without losing access to native provider features.
Supercompat vs. LiteLLM
LiteLLM has two deployment modes:
Python SDK (
from litellm import completion) — runs in-process in Python apps. Unifies providers behind LiteLLM's API. Python-only on the client.LiteLLM Proxy — a self-hosted OpenAI-compatible gateway server. Your app talks to it over HTTP in OpenAI format. This is the only supported integration path for JS/TS apps; there's no official LiteLLM JS SDK.
Supercompat runs in-process in TypeScript/JavaScript. Single npm dependency, no proxy server, no HTTP hop, no separate deployment story. Your app imports
openai or @anthropic-ai/sdk and uses it natively.Pick LiteLLM if you're on Python, or if you want a central gateway with its dashboard for auth, cost tracking, and rate-limiting.
Pick Supercompat if you're on TypeScript and want to eliminate the proxy hop while keeping the full native SDK surface.
Supercompat vs. Portkey / Helicone
Portkey and Helicone are observability / routing proxies — their primary value is caching, analytics, fallbacks, key management, and usage tracking. You send OpenAI-format requests to them; they forward and log.
Supercompat is a client-side library. It has no observability dashboard and no cloud component. It's focused on turning one SDK into any provider in-process.
The two compose. You can use Supercompat to get an OpenAI client pointed at any provider, then set its
baseURL to Portkey or Helicone to layer observability on top.Supercompat vs. OpenRouter (direct)
OpenRouter is a backend provider (an HTTPS endpoint that exposes 300+ models via the OpenAI wire format). Calling OpenRouter directly with the OpenAI SDK already works for basic chat completions.
Supercompat + OpenRouter (via
openRouterClientAdapter) adds:The Responses API and Assistants API on top of OpenRouter's chat-completions endpoint.
Storage adapters so you get persisted threads, runs, conversations.
Provider routing hints (
provider: { order, allow_fallbacks }) forwarded on every request.Free switching between OpenRouter and a direct provider when you want to stop going through an aggregator — same code.
Why choose Supercompat
Real SDK types, full provider features. You keep every capability the provider exposes — not a lowest-common-denominator abstraction.
Responses + Assistants surfaces against any provider. Not just chat completions.
In-process. No proxy, no sidecar, works in edge runtimes.
Composable adapters. Pick one client × one run × one storage; swap any layer without touching the rest of your code.
Storage you own. Prisma + Postgres, memory, OpenAI-managed, Azure-managed — your call.
When Supercompat is not the right fit
You're on Python. (Use LiteLLM's Python SDK.)
You want a framework with agents, chains, memory modules, retrievers out of the box. (Use LangChain or Vercel AI SDK.)
You want a hosted gateway for observability, caching, or centralized key management. (Use Portkey, Helicone, or LiteLLM Proxy.)
You only use OpenAI, on OpenAI, forever. (Just use
openai directly.)