RelayPlane vs Portkey: Which LLM Proxy is Right for Node.js?
Portkey open-sourced its Go AI Gateway in April 2026, claiming 2 trillion tokens per day at scale. Both projects are now MIT licensed. The real difference is architecture: Go gateway with an npm SDK wrapper vs. a pure Node.js proxy with a single npm install, local-first storage, and an instant kill-switch.
TL;DR
Choose RelayPlane when you want:
- Pure Node.js with a single npm install
- Local-first cost tracking with no external service
- Ollama and local model fallback routing
- Budget tracking without a cloud dashboard
Choose Portkey when you need:
- A Go-native gateway at massive scale (2T+ tokens/day)
- Multi-cloud guardrails with the Portkey hosted control plane
- Virtual keys and team analytics via the hosted dashboard
Feature Comparison
| Feature | RelayPlane | Portkey |
|---|---|---|
| Install method RelayPlane is one npm command, self-contained. Portkey splits across a Go gateway binary and a Node.js SDK wrapper. | npm install -g @relayplane/proxy | npm install @portkey-ai/gateway (Node SDK) + separate Go gateway binary |
| Language / runtime RelayPlane runs natively in Node.js with no additional runtimes. Portkey AI Gateway is written in Go; the npm package is a thin SDK over it. | Node.js / TypeScript native | Go gateway + Node.js SDK |
| External service dependency RelayPlane is fully self-contained. Portkey can run fully self-hosted but requires standing up its Go gateway separately. | None | Optional hosted gateway or self-hosted Go binary |
| Local-first (no cloud account required) RelayPlane works entirely offline with no cloud account. Portkey self-hosted skips the cloud, but many features (virtual keys, analytics) require the hosted service. | Partial | |
| Per-request cost tracking RelayPlane tracks exact token costs per request in local SQLite with no external service. Portkey cost tracking ties to their hosted dashboard or requires operating their analytics stack. | Via hosted dashboard or self-hosted analytics | |
| Ollama / local model fallback RelayPlane natively routes to Ollama and local models as a fallback. Portkey AI Gateway does not support Ollama or local model routing. | ||
| Smart routing by complexity RelayPlane routes to Haiku, Sonnet, or Opus based on task complexity automatically. Portkey requires manual routing rule configuration. | ||
| Data stays on your machine RelayPlane stores all telemetry and cost data locally in SQLite. Portkey hosted sends data to their cloud; self-hosted avoids this but requires operating the full stack. | Depends on setup | |
| Open source license Both are MIT licensed. Portkey fully open-sourced its AI Gateway in April 2026, claiming 2 trillion tokens/day throughput. The hosted Portkey platform still offers managed cloud features, but the core gateway code is now fully open. | MIT | MIT (fully open-sourced April 2026) |
| Claude Code / Cursor compatible RelayPlane is designed and tested with Claude Code, Cursor, and OpenClaw. Portkey works as a generic proxy but is not specifically optimized for these tools. | Partial | |
| Pricing RelayPlane is fully free and open source. Portkey hosted has a free tier with rate limits; enterprise features require paid plans. | Free (open source) | Free tier + paid plans |
| Per-tenant isolation RelayPlane's tenant isolation runs at the proxy layer with in-process enforcement, no cloud account required. Portkey's virtual key system provides per-key budget tracking but requires their hosted service. | First-class: separate request queues, budgets, and audit namespace per tenant | Virtual keys provide per-key budget limits (cloud-based) |
| Instant kill-switch / intervention API RelayPlane provides a single HTTP call that stops all traffic for a tenant in-process, no round-trip to a cloud control plane. Portkey's open-source gateway supports rate limits and guardrails but has no instant kill-switch that works without the hosted control plane. | POST /v1/tenants/:id/kill, halts all traffic within one request cycle | No equivalent (rate limits and guardrails only) |
| Spec-match verification RelayPlane's spec-match plugin verifies agent outputs against acceptance criteria before task completion. Portkey has no spec verification capability. | Built-in: LLM evaluates agent output against acceptance criteria | |
| Compliance audit bundle export RelayPlane's audit chain uses cryptographic checksums so tampering is detectable. Portkey provides request logs in its hosted dashboard with enterprise audit features on paid tiers. | Tamper-proof HMAC-chained audit log, exportable as JSON/CSV/JSONL | Request logs via hosted dashboard; enterprise audit logging on paid plans |
Why Node.js Developers Choose RelayPlane
One npm install, zero extra runtimes
npm install -g @relayplane/proxy and you are running. No Go binary to compile, no separate gateway process to operate, no external service to sign up for. It fits directly into any Node.js workflow.
Per-request cost tracking without leaving your machine
RelayPlane tracks exact token costs per request in self-hosted SQLite. No dashboard to pay for, no telemetry leaving your server, no external dependencies. See exactly what every request costs without any infrastructure overhead.
Ollama and local model fallback built in
Route to Ollama or any local model when cloud providers are slow or too expensive. Portkey AI Gateway has no local model support. RelayPlane treats local models as first-class routing targets.
Smart complexity routing with no manual rules
RelayPlane automatically routes simple tasks to Haiku, moderate tasks to Sonnet, and complex tasks to Opus. No routing rules to write and maintain. Portkey requires you to configure routing logic manually.
Portkey Open-Sourced in April 2026. The Architecture Difference Still Matters.
Portkey open-sourced its AI Gateway in April 2026, reporting 2 trillion tokens per day throughput. It is now MIT licensed and self-hostable without paying for a cloud plan. That is a meaningful shift for developers who previously ruled it out on cost grounds.
The remaining differentiators are architectural. Portkey is a Go binary with a Node.js SDK wrapper, which means running it locally requires a separate gateway process. RelayPlane is a single npm package, pure Node.js, with no additional runtimes. Cost data stays in local SQLite. Ollama and local model fallback work out of the box.
The other gap is intervention capability. RelayPlane ships a kill-switch API that halts all tenant traffic in one in-process request, no round-trip to a cloud control plane. That matters for agentic workloads where a stuck or misbehaving agent needs to be stopped immediately. Portkey's open-source gateway supports rate limits and guardrails but has no equivalent single-call kill-switch that works fully offline.