Live · Open for demo

Meridian

One gateway for OpenAI and Anthropic. Less cost, less risk.

Meridian cost insight: recent traffic repriced on every model, and requests per day split between provider and cache Meridian developer view: an OpenAI SDK client pointed at the gateway, with the feature list

Overview

Meridian sits between your applications and AI providers. It checks each caller's key, limits, and budget, picks a provider with fallback, redacts personal data, caches where safe, and records the cost of every request.

Existing OpenAI and Anthropic SDK code — and Claude Code — works by changing the base URL and the key.

Open for demo: request access at app.sdwares.com. Approved workspaces start with free trial credit, then bring their own provider keys.

At a Glance

1 URL Change to adopt
2 Native APIs: OpenAI + Anthropic
5 Roles per organization
100% Requests cost-attributed

Key Features

⬡

Drop-In API

OpenAI Chat Completions, the native Anthropic Messages API, embeddings, and an MCP server for Claude Desktop and other MCP clients.

◈

Budgets That Hold

Monthly budget per project, with cost reserved before each call so parallel requests can't overshoot. Email alerts at 80% and 100%.

↻

Failover & Aliases

Model aliases route one name across providers. Retries with backoff and a circuit breaker per provider and model.

⚡

Exact & Semantic Cache

Exact-match cache for deterministic requests, optional semantic cache for similar ones. Both switchable per project.

◎

PII Redaction & BYOK

Personal data redacted before it reaches a provider, streams included. Bring your own keys, envelope-encrypted with Azure Key Vault.

▤

Self-Service Portal

Organizations, projects, keys, team roles, usage charts, request and audit logs, and a "what if" simulator that reprices your traffic on every model.

Architecture

Node.js 24 · TypeScript · Fastify 5 PostgreSQL · Prisma Redis Next.js on Azure Static Web Apps Azure Container Apps Azure Key Vault Terraform GitHub Actions (OIDC) OpenTelemetry · New Relic · Langfuse

Key Decisions

  • →Container Apps over Kubernetes — no cluster to run; trade some control for a much smaller ops surface
  • →In-process format translation — no extra network hop; native Anthropic pass-through for full fidelity
  • →Cookie sessions, static dashboard — the browser never holds a token, and the dashboard needs no server
  • →Cost reservations — budgets and trial credit hold under parallel load
  • →Expand-only migrations — run before each rollout, so the previous revision always keeps working

Interested in AI infrastructure
for your team?

Try Meridian in the demo, or let's talk about something purpose-built for your organization.

Try the Demo ↗