AI gateway

One model endpoint. Every routing decision under your control.

Put one OpenAI-compatible API in front of every provider, then centralize routing, fallback, caching, limits, credentials, and tenant boundaries.

Start with $5 credit Read the docs

One-time starter credit per account on Everstack Cloud. Expires 30 days after signup.

01OpenAI-compatible API
02200+ catalog models
03Fallback and balancing
04Tenant-scoped controls

01Operating model

One path from application intent to controlled execution.

Put one OpenAI-compatible API in front of every provider, then centralize routing, fallback, caching, limits, credentials, and tenant boundaries.

Keep application code stable

Your application sends one request shape to one endpoint. Everstack translates provider formats and keeps model changes out of product code.

Route with operational context

Select providers through aliases, explicit routes, catalog metadata, health, and configured balancing strategies instead of hard-coded SDK calls.

Carry policy through every request

Authentication, tenant isolation, cache scope, rate limits, and telemetry stay attached from ingress to the upstream provider response.

02Capabilities

The controls needed to run it in production.

Everstack keeps gateway routing connected to the same identity, policy, deployment, and telemetry boundaries as the rest of the platform.

Provider abstraction

Translate OpenAI-shaped requests into provider-native calls while preserving streaming, tools, structured output, and provider-specific reasoning state.

Fallback chains

Move traffic to a configured alternative when a provider is unavailable, rate-limited, or outside the policy you set.

Exact and semantic caching

Serve repeated requests without another provider call, with tenant-scoped entries, configurable similarity, and explicit TTLs.

Rate limiting

Enforce request budgets per tenant and per key before traffic reaches a model provider.

Context compaction

Trim conversation history when it exceeds the selected model's context window while preserving the request contract.

Traceable economics

Capture model selection, cache decisions, latency, token use, cost, and fallback behavior in the same request trace.

03Build with your agent

Give the agent the integration contract, not just a code fragment.

Choose the client that fits the project, inspect the highlighted reference, or copy a complete implementation brief with credential, routing, and verification requirements included.

Agent-ready integration

Route a model call through Everstack

Start with an Everstack SDK, keep an existing compatible client, or give the complete implementation brief to a coding agent.

generate.ts

import Everstack from "@everstack/node";const everstack = new Everstack({  apiKey: process.env.EVERSTACK_API_KEY!,  baseUrl: process.env.EVERSTACK_GATEWAY_URL,});const stream = await everstack.chat.completions.create({  model: "@google/gemini-3.6-flash",  messages: [{ role: "user", content: "Why is the sky blue?" }],  stream: true,});for await (const chunk of stream) {  process.stdout.write(chunk.choices[0]?.delta?.content ?? "");}

Read the docs

Project aware: The brief tells the agent to reuse the client and conventions already present.
Secret safe: Credentials stay in environment variables and out of source, logs, and browser code.
Evidence included: The task finishes only after a real request and its Everstack trace are verified.

04Where it fits

Built for real workloads, not isolated demos.

Start with the narrowest useful surface, then connect more of the platform as the workload needs stronger execution, policy, or evidence.

Use case 01

Multi-provider products

Keep one integration while choosing the best provider for each workload.

Use case 02

Resilient production traffic

Keep requests moving through upstream incidents and rate limits.

Use case 03

Cost and latency control

Combine routing, caching, limits, and request-level telemetry.

05Questions

What teams usually ask first.

A direct view of how this surface behaves, where it runs, and how it connects to the rest of Everstack.

Do applications need a new SDK?

No. Existing OpenAI-compatible clients can point at the Everstack gateway by changing the base URL and credentials.

Can each tenant have different providers?

Yes. Provider credentials, aliases, cache entries, and limits can remain isolated inside each tenant boundary.

Can the gateway run without the rest of the platform?

You can start with routing alone, then enable caching, rate limiting, guardrails, and observability as your operating requirements grow.

Reference

The pages you open once you start building.

Continue through the stack

Next, explore agents + workflows.

Open the product page