AI gateway and observability · Verified August 16, 2026

TrustedRouter vs Portkey

Portkey pairs a gateway with the most complete single-vendor observability stack in the category — and, since May 2026, Palo Alto Networks ownership; its hosted platform logs full request content by default to power that product. TrustedRouter routes realtime traffic with no durable prompt logs and a hardware-attested gateway build you can verify.

Compare details
8decision dimensions
8official sources
1base URL to try TR
Choose Portkey when

Its operating model is the feature.

Choose Portkey when prompt-level logs, traces, guardrails, and prompt management are your operating model: it is the most complete single-vendor stack in this category, with a free tier, a $49/month Production plan, SOC 2 Type 2 and HIPAA at the Enterprise tier, a hybrid VPC deployment mode, and Palo Alto Networks behind it.

Choose TrustedRouter when

Privacy needs evidence.

Choose TrustedRouter when prompt content must not persist anywhere on the gateway and you want that checkable rather than contractual: realtime routing with no durable prompt or output logs, privacy-tier routes (zdr, e2e, eu), and a gateway build verified against live attestation on three clouds.

DimensionPortkeyTrustedRouter
DeploymentSelf-host MIT gateway, hosted SaaS, or hybrid: VPC data plane + hosted control planeHosted control plane, source-available (BUSL-1.1), with an attested API path
API surfaceOpenAI-compatible unified API via SDK or x-portkey-* headers; 40+ provider adaptersOpenAI Chat Completions and Responses plus Anthropic Messages
Model accessClaims 1,600+ LLMs across 40+ providers; 50+ guardrails500+ models across ~50 providers: prepaid, BYOK, and direct
RoutingFallbacks, load balancing, retries, conditional routing, simple + semantic cachingProvider fallback plus auto, cheap, fast, free, ZDR, E2E, and EU routes with composable privacy and jurisdiction preferences
ObservabilityLogs, traces, metrics, alerts, dashboards; plans meter recorded logsMetadata analytics and opt-in external broadcast
Prompt contentLogs full content by default; org-wide metrics-only mode is Enterprise-onlyNo durable prompt or output logs on realtime inference; batch is opt-in encrypted retention
VerificationSOC 2 Type 2, ISO 27001, HIPAA audits; config + contract, no runtime attestationLive gateway attestation on three clouds, bound to published source and release evidence
BillingFree 10k logs/mo; $49 for 100k + $9/extra 100k; inference on your own provider keysPrepaid at provider price + 5.5% ($0.01/M floor), or BYOK

What Portkey is in 2026

Portkey is a gateway built around an observability platform: request logs, traces, analytics, 50+ guardrails (their count), prompt management, governance, and an MCP gateway from one vendor. It ships in two parts. The gateway is MIT-licensed TypeScript with 12,740 GitHub stars as of August 16, 2026; its README claims routing to 1,600+ LLMs across 40+ providers. The hosted platform — log store, dashboards, prompt studio, RBAC — is proprietary, and it is what Portkey's pricing page sells.

The most material fact about Portkey in 2026 is ownership. Palo Alto Networks announced intent to acquire on April 30, 2026 and completed the acquisition on May 29, 2026; Portkey is now the foundational AI gateway of Prisma AIRS, Palo Alto's AI security platform. In March 2026 Portkey announced a $15M Series A and a fully open-source Gateway 2.0. As of August 16, 2026 that 2.0 work sits on a pre-release branch with no root LICENSE file, and the main branch has had no pushes since May 25, 2026. The commercial product is active; the post-acquisition open-source cadence is unproven.

Scale, company-reported in March 2026: 24,000+ organizations, 1 trillion+ tokens and 120M+ requests processed daily. We could not verify those figures independently, and we know of no reason to doubt them.

Where Portkey is genuinely strong

  • One vendor, whole stack. Teams that debug from prompt-level logs and traces, gate outputs with guardrails, and version prompts in a studio get all of it behind one API. TrustedRouter does none of that; our observability is metadata-only by design.
  • Cheap entry. A free Developer tier (10k recorded logs per month) and a $49/month Production tier (100k logs, $9 per additional 100k) buy the working platform.
  • Audited compliance. Portkey's security documentation lists SOC 2 (Type 2 at Enterprise), ISO 27001, GDPR, and HIPAA with third-party audits. We publish a DPA, BAA, and subprocessor list, and no third-party certification today.
  • Hybrid VPC deployment. Enterprise customers run the data plane — gateway and log store — inside their own network; Portkey's deployment docs state LLM traffic stays inside your network boundary in that mode. The control plane stays Portkey-hosted, and configs, including provider API keys, sync to it.

What happens to your prompts

This is the load-bearing difference, so we will be precise. Portkey's hosted observability works because it stores content: the logs documentation describes a chronological view of every request processed through the gateway, request and response bodies included, retained 3 days on the free tier and 30 days on Production. Opting out is per-request (x-portkey-debug: false), and the org-wide Metrics Only mode is an Enterprise plan feature. Switching modes is not retroactive — content already logged stays until support purges it on request.

TrustedRouter is built in the opposite order. Realtime inference never writes prompt or output content to durable storage, on any plan. What we keep is metadata: request and generation ids, model and provider, token counts, latency, status, cost, region, and an API-key hash (privacy page). Batch is a separate opt-in mode with enclave-encrypted retention up to 30 days. Teams that want content-level analytics can enable Broadcast to send selected content to their own destination; we do not retain a copy.

The second half of the difference is proof. Portkey's privacy posture is configuration plus contract; nothing in its docs attests that a given logging mode is in effect on the serving path. TrustedRouter publishes live hardware attestation for the gateway on three clouds — GCP Confidential Space, AWS Nitro Enclaves, and Azure Confidential Containers — plus a verifier script that checks the running enclave against published source and release digests. Start at trust.trustedrouter.com and /security.

Fees on different bases

Do not compare the numbers directly; the meters differ. Portkey prices recorded logs — free at 10k per month (the pricing page itself calls that tier unsuitable for production), $49/month for 100k, $9 per additional 100k, custom at Enterprise — while model inference bills separately through your own provider keys. TrustedRouter has no subscription: prepaid text and embeddings bill the provider's token price plus 5.5% with a $0.01 per million token floor, video is the provider quote plus 20%, and every route's price is listed on /models (pricing details). BYOK is supported; we do not publish a separate BYOK fee. One meter charges for observability volume, the other marks up tokens. Which is cheaper depends on your traffic shape.

What we do not claim

  • Attestation covers our gateway, not model providers. Downstream handling is tracked policy and contract per provider, except trustedrouter/e2e, which routes only to providers running confidential compute with provider-side end-to-end encryption (Tinfoil, Phala) — their mechanisms, not our attestation.
  • Our repos are young: public since late April and early May 2026. Portkey's MIT gateway has 12,740 stars and 1,247 forks; ours has months of public history.
  • Our gateway and control plane are source-available under BUSL-1.1, converting to Apache-2.0 four years after each release; the SDKs are Apache-2.0 and MIT. Portkey's gateway on main carries the more permissive license. Every line of ours that touches your prompt is public.
  • No published SOC 2 or HIPAA certification today. If audited certification is a hard requirement now, Portkey Enterprise has it and we do not.

Migration reality

Plain proxy usage moves in minutes: point the OpenAI SDK at the TrustedRouter base URL and replace x-portkey-* headers with one key. The real work is elsewhere. Portkey Configs — fallback, load-balancing, retry, and cache JSON — must be re-expressed as TrustedRouter routing preferences: provider ordering, models[] fallback arrays, and privacy floors. Prompt templates live in Portkey's hosted Prompt Studio and render through its API, so export them before you leave. Guardrail hooks have no TrustedRouter equivalent and need re-implementation in your application. Logged history stays behind under Portkey's retention windows. We have no Portkey-specific importer; our only dedicated migration guide today covers OpenRouter.

Migration shape

Start with one real request.

Point your OpenAI-compatible client at the TrustedRouter base URL and replace x-portkey-* headers with one TrustedRouter key. Re-express Portkey Configs (fallbacks, load balancing, retries, caching) as TrustedRouter routing preferences where equivalents exist — provider ordering, models[] fallbacks, privacy floors, and export prompt templates from Prompt Studio first — they are hosted assets and do not carry over. Guardrail hooks and logged history do not migrate.

Keep the first test small, stream the response, and compare output, latency, provider selection, and billed usage before moving production traffic.

Agent setup Run a small eval

TrustedRouter sideOpenAI SDK
from openai import OpenAI

client = OpenAI(
    base_url="https://api.trustedrouter.com/v1",
    api_key="sk-tr-v1-...",
)

with client.chat.completions.create(
    model="trustedrouter/zdr",
    messages=[{"role": "user", "content": "Reply PONG"}],
    stream=True,
) as response:
    for chunk in response:
        print(chunk.choices[0].delta.content or "", end="")
Official evidence

Sources checked August 16, 2026

Report a change
All gateways

Questions

Can TrustedRouter replace Portkey's observability, guardrails, and prompt management?

Mostly no, and we want to be plain about it. TrustedRouter keeps metadata only — ids, model, tokens, latency, cost, region — with opt-in Broadcast export for content you choose to send to your own destination. There is no hosted log viewer with prompt bodies, no trace UI, no guardrails engine, no prompt studio. If prompt-level debugging and guardrails are central to how your team works, Portkey is the stronger product for that job.

Is Portkey's Metrics Only mode equivalent to TrustedRouter's no-log design?

No. Portkey's org-wide Metrics Only mode is an Enterprise-plan feature; lower tiers log full content by default with a per-request opt-out header, and switching modes does not purge already-logged data. It is also a configuration you trust rather than verify. TrustedRouter's realtime path never durably stores prompt or output content on any plan, and the gateway build serving that promise is hardware-attested on three clouds with a public verifier at trust.trustedrouter.com.

Does the Palo Alto Networks acquisition change the calculus?

It depends on your stack. If you already run Palo Alto security products, Portkey as the gateway core of Prisma AIRS is a real integration advantage. If you rely on the open-source gateway, note that as of August 16, 2026 the main branch has had no pushes since May 25, 2026 and the last tagged release was January 2026; the roadmap is now oriented around Prisma AIRS. Self-hosting the MIT gateway remains viable today either way.

Workspace access

Sign in

Choose a sign in method to access your TrustedRouter workspace.

By signing in you agree to the terms of service and privacy policy.