Alternatives / Vercel AI Gateway
Looking for an open source Vercel AI Gateway alternative? We compared 5 open source LLM gateways by license, routing, guardrails, observability, and maintenance. LiteLLM is the most widely used, while Portkey, Bifrost, Agent Router, and LLM Gateway take different approaches.
Written by Maxime DERAME
Last updated:
| Alternative | Best for | License |
|---|---|---|
| LiteLLM | Proxy for 100+ models with budgets | MIT + enterprise directory |
| Portkey AI Gateway | Gateway with guardrails and 1,600+ models | MIT |
| Bifrost | High-performance Go gateway | Apache-2.0 |
| Envoy AI Gateway (Agent Router) | Envoy-based gateway for agent traffic | Apache-2.0 |
| LLM Gateway | OpenAI-compatible gateway with analytics | AGPL-3.0 + EE |
We looked for open-source gateways that route requests to many model providers through one API like Vercel AI Gateway. We verified each project's license, recent activity, stars, and self-hosting in its repository. We left out one gateway whose core repository we could not verify. See our methodology for the full criteria.
LiteLLM is a proxy for 100+ LLMs in the OpenAI format with virtual keys, budgets, rate limits, fallbacks, and cost tracking, self-hostable with Docker. Portkey's gateway routes to 1,600+ LLMs with guardrails and fallbacks.
Bifrost is an Apache-2.0 gateway with 1,000+ models, governance, fallback, and OpenTelemetry support, with its latency figures self-reported. Agent Router, formerly Envoy AI Gateway, is built on Envoy and is aimed at teams already running Envoy.
LLM Gateway routes and analyzes requests across providers through an OpenAI-compatible API, with free self-hosting. Its hosted free plan adds a 5% gateway fee.
LiteLLM's SSO, JWT auth, audit logs, and custom SLAs need its enterprise tier. LLM Gateway has enterprise-licensed parts, and Portkey's governance features are in its hosted platform. Bifrost's enterprise features such as SSO require a trial or license.
LiteLLM is MIT with an enterprise directory. Portkey is MIT. Bifrost and Agent Router are Apache-2.0. LLM Gateway is AGPL-3.0 with enterprise parts.
LiteLLM, Bifrost, Agent Router, and LLM Gateway had commits within days of this update. Portkey's last commit was in May 2026.
Vercel AI Gateway is a proprietary hosted routing service, so your prompts pass through Vercel and billing follows its rates. Open-source gateways can run in your own infrastructure, so you keep provider keys, apply your own budgets and guardrails, and avoid platform markups.
LiteLLM is an AI gateway and Python SDK that calls 100+ LLM APIs in the OpenAI format, with a proxy server that adds virtual keys, spend tracking and budgets per team or user, rate limiting, load balancing, automatic fallbacks, guardrails, and observability integrations such as Langfuse and OpenTelemetry. The repository is MIT outside its enterprise directory, which adds SSO, audit logs, and other features under a separate license.
Pricing: Free open-source proxy; paid enterprise tier and hosted proxy.
Portkey AI Gateway is an MIT gateway that routes requests to 1,600+ LLMs through one API with guardrails, fallbacks, and caching. The open-source gateway is one part of the Portkey platform, whose observability and governance features are hosted. The last commit was in May 2026.
Pricing: Free open-source gateway; platform features are paid.
Bifrost is a high-performance AI gateway that gives fast access to a large catalog of models from many providers, with an adaptive load balancer, virtual keys with independent budgets, real-time guardrails, an MCP gateway for tool management, SSO, and OpenTelemetry support. Its maintainers claim it is much faster than LiteLLM.
Pricing: Free and open source; enterprise features available with a free trial.
Envoy AI Gateway, now renamed Agent Router, is an Apache-2.0 community project built on Envoy Gateway that routes AI and agent traffic to many providers through an OpenAI-compatible API. It targets platform teams already running Envoy and does not offer a built-in cost dashboard.
Pricing: Free and open source.
LLM Gateway routes LLM requests across multiple providers through one OpenAI-compatible API, with usage analytics, performance monitoring, and secure key management, and SDKs for languages such as Python, TypeScript, Java, Rust, and Go. The core is AGPL-3.0, enterprise features in the ee directory are under a commercial license, and a hosted service is available.
Pricing: Free to self-host; hosted free plan with a 5% gateway fee, Pro at $50 per month, and custom enterprise plans.
| Alternative | Type | License | Self-hosted | Best for |
|---|---|---|---|---|
| LiteLLM | LLM gateway | MIT + enterprise directory | Yes | Closest replacement |
| Portkey AI Gateway | LLM gateway | MIT | Yes | Guardrails and fallbacks |
| Bifrost | LLM gateway | Apache-2.0 | Yes | Low-latency routing |
| Envoy AI Gateway (Agent Router) | LLM gateway | Apache-2.0 | Yes | Envoy and Kubernetes teams |
| LLM Gateway | LLM gateway | AGPL-3.0 + EE | Yes | Usage analytics |
Want the most widely used gateway
LiteLLM supports 100+ models with virtual keys, budgets, and observability integrations.
Want guardrails built in
Portkey's gateway includes guardrails and routes to 1,600+ models.
Want very low overhead
Bifrost is Go-based with Apache-2.0 licensing and OpenTelemetry support.
Run Envoy on Kubernetes
Agent Router builds on Envoy Gateway and routes AI and agent traffic.