Comparison · June 2026
SpendProxy vs Portkey
Portkey is an AI gateway and observability/governance platform with a unified API to 1,600+ models — routing, fallbacks, caching (simple and semantic), guardrails, and prompt management. The gateway is open source (MIT). Portkey was acquired by Palo Alto Networks (closed May 2026) and is becoming the AI gateway of PANW’s Prisma AIRS security platform.
What Portkey is genuinely good at
- ✓Very broad model coverage (1,600+) behind one OpenAI-compatible API
- ✓Mature reliability primitives: fallbacks, retries, load balancing, canary
- ✓Semantic caching available at an inexpensive tier ($49/mo)
- ✓Hybrid deployment: data plane in your VPC with SaaS control plane
Side by side
| Portkey | SpendProxy | |
|---|---|---|
| Primary job | Production gateway + governance for Gen-AI apps | Cut LLM spend: billing-accurate cost tracking plus optimization engines that act on traffic |
| Request path | Inline gateway (SaaS, hybrid, or self-hosted OSS) | Inline proxy — one base-URL change |
| Hosting & data path | SaaS routes traffic through Portkey cloud; hybrid/Enterprise keeps LLM traffic in your VPC with a SaaS control plane | Licensed Docker container inside your VPC. SQLite-local storage. No request data leaves your network. |
| Automatic cost optimization | Config-driven: routing rules are written by you; provider prompt caching (e.g. cache_control) must be added by you — Portkey tracks it but does not inject it | Five engines that take action: prompt cache injection, response dedup, model routing with circuit breaker, budget guardrails, retry-storm suppression. Each runs in off, monitor (log what it would do), or autopilot mode. |
| Cost tracking | Request logging with cached-token pricing tracked | Provider-specific billing semantics: cached tokens, reasoning tokens, and streaming priced the way the provider actually bills. |
| Budgets & enforcement | Cost/token budgets with auto-block — gated to Enterprise and select Pro plans | Per route, tag, or API key — warn or hard-block. |
| Provider coverage | 1,600+ models | OpenAI, Anthropic, Google Gemini. Deliberately deep on three providers rather than broad. |
| Source / license | Gateway open source, MIT; platform commercial | Commercial. Licensed, air-gapped container. |
| Pricing | Free (10K logs) · Production $49/mo · Enterprise custom | $2,500 30-day pilot, then $1,500/mo flat. |
Choose Portkey when
- →You want maximum model/provider breadth with reliability features
- →You want semantic caching cheaply and are fine configuring it
- →A Palo Alto Networks-backed security roadmap is a plus for you
Choose SpendProxy when
- →You want optimization to happen without writing and maintaining routing configs — monitor mode shows the savings, autopilot takes them
- →Budget enforcement on day one, not behind an enterprise gate
- →Fully air-gapped: no SaaS control plane, nothing leaves the VPC
See it on your own traffic
30-day pilot inside your VPC. Monitor mode shows exactly what each engine would save before anything changes. If the numbers aren't there, you'll know in week one.
Portkey facts verified June 2026 against their official documentation. If anything here is out of date, email hi@spendproxy.com and we'll fix it.