LLM observability and gateway · Verified August 16, 2026

TrustedRouter vs Helicone

Helicone is an Apache-2.0 LLM observability platform with an AI gateway, in maintenance mode since Mintlify acquired it on March 3, 2026. TrustedRouter is an actively developed model router whose realtime path keeps no durable prompt logs and whose gateway build is attested live on three clouds.

Compare details
8decision dimensions
7official sources
1base URL to try TR
Choose Helicone when

Its operating model is the feature.

Choose Helicone when request-level observability is the job: full-body logs, session traces, HQL, prompt management, and evals, with a real free tier and an Apache-2.0 monorepo you can self-host. Weigh that against maintenance mode: since the March 2026 Mintlify acquisition, feature development has stopped.

Choose TrustedRouter when

Privacy needs evidence.

Choose TrustedRouter when you want one prepaid API across 550+ model routes (as of August 2026) where ordinary synchronous and streaming inference keeps no durable prompt or output logs, with a gateway build you can verify by live attestation on three clouds. You accept metadata-only observability in exchange.

DimensionHeliconeTrustedRouter
DeploymentHosted (US/EU) or self-hosted Apache-2.0 platform; maintenance mode since Mar 2026Hosted control plane, source-available (BUSL-1.1), with an attested API path
API surfaceOpenAI-compatible cloud gateway, per-provider passthrough domains, or async log SDKsOpenAI Chat Completions and Responses plus Anthropic Messages
Model access100+ models via cloud gateway; pass-through billing credits or BYOK provider keys500+ models across ~50 providers: prepaid, BYOK, and direct
RoutingLoad balancing, automatic provider fallback, caching, and custom rate limitsProvider fallback plus auto, cheap, fast, free, ZDR, E2E, and EU routes with composable privacy and jurisdiction preferences
ObservabilityFull-body request logs, sessions, users, HQL, prompts, datasets, evals, alertsMetadata analytics and opt-in external broadcast
Prompt contentBodies logged by default; omit headers skip storage but content still transitsNo durable prompt or output logs on realtime inference; batch is opt-in encrypted retention
VerificationApache-2.0 source and self-hosting; documented SOC 2; SOC 2 & HIPAA plan features on Team ($799/mo) and upLive gateway attestation on three clouds, bound to published source and release evidence
BillingPlans $0-$799/mo plus usage-based overage; credits priced at 0% markup (waitlist)Prepaid at provider price + 5.5% ($0.01/M floor), or BYOK

What Helicone is in August 2026

Helicone is an LLM observability platform with a gateway attached. The core is real open source: the main monorepo is Apache-2.0 with about 6,000 GitHub stars, and you can run it self-hosted or use the hosted US and EU regions. At acquisition the team reported 16,000 organizations and 14.2 trillion tokens processed over three years. That is proven scale, and the one-line integration that made it popular still works.

The acquisition is where any 2026 evaluation has to start. On March 3, 2026, Mintlify acquired Helicone. Helicone's own announcement says services stay live "in maintenance mode": security updates, new model support, and bug fixes continue, and Mintlify offers to help customers migrate to another platform. Maintenance mode means feature development has stopped. The public record matches. The changelog's last entry is November 26, 2025, and the standalone Rust ai-gateway repo has been idle since November 21, 2025 — its final commit relicensed it from Apache-2.0 to GPL-3.0.

Where Helicone is genuinely strong

Request-level observability is Helicone's home turf, and it is better at it than we are. One integration line gets you full request and response bodies, session traces, per-user analytics, cost breakdowns, an SQL-like query language (HQL), prompt management, datasets, evals, and a playground. TrustedRouter keeps metadata-only analytics by design; we do not offer a comparable debugging surface.

The pricing is transparent and generous: a free tier of 10,000 requests per month, Pro at $79/month with unlimited seats, usage-based overage pricing, and startup and student discounts. The cloud gateway's pass-through billing credits are priced at 0% markup with only payment processing fees, though as of August 2026 the credits page still gates access behind a waitlist. Because the platform is Apache-2.0, self-hosting is a genuine escape hatch, with on-prem deployment on the Enterprise plan.

Three differences that decide it

Logging by default, or no durable logs

Helicone's premise is capturing prompts and completions; bodies are logged by default, since that is what the dashboards, evals, and HQL run on. Per-request opt-outs exist (Helicone-Omit-Request and Helicone-Omit-Response), and they do stop storage, but Helicone's own docs note the content is still sent to their backend. Keeping payloads out entirely requires async logging with content tracing off, which takes the gateway out of the request path. Retention is plan-gated: 7 days free, 1 month at $79, 3 months at $799.

TrustedRouter inverts the default. Ordinary synchronous and streaming prompt paths do not touch persistent storage: no durable prompt or output logs. We retain operational metadata — request ids, model and provider, token counts, latency, cost, region, API-key hash. Batch is a separate opt-in mode with enclave-encrypted retention up to 30 days. Exact wording is on our privacy page.

How you verify the privacy claim

Helicone's trust model is source plus policy: Apache-2.0 code you can self-host, and documented SOC 2 compliance, with "SOC-2 & HIPAA compliance" sold as a feature of the $799/month Team plan and above. TrustedRouter's is different in kind: api.trustedrouter.com runs inside trusted execution environments on GCP Confidential Space, AWS Nitro Enclaves, and Azure Confidential Containers, and live attestation endpoints at trust.trustedrouter.com bind the running gateway to published source and release digests. The boundary, stated plainly: attestation covers our gateway build. Downstream model providers are covered by cited, hand-audited retention policy tiers, except the trustedrouter/e2e routes, which only reach providers running their own confidential compute with end-to-end encryption. See /security.

Fees on different bases

Do not compare the fee numbers directly; the bases differ. Helicone charges for the platform: plan fees from $0 to $799/month plus usage-based per-request overage, while its gateway credits are priced at 0% markup on inference itself (waitlist-gated as of August 2026). TrustedRouter charges no plan or seat fee; prepaid text and embeddings bill the provider's token price plus 5.5%, with a $0.01 per million token floor. BYOK is supported, and we publish no separate BYOK fee. Per-route prices are listed on /models and full terms on /pricing.

What we do not claim

  • Our gateway and control plane are source-available under BUSL-1.1, converting to Apache-2.0 four years after each release. Helicone's monorepo is OSI open source today. Every line that touches your prompt is public; our SDKs are Apache-2.0 or MIT.
  • We publish no SOC 2 or HIPAA certification as of August 2026. Helicone documents SOC 2 compliance.
  • Our repos have been public since late April 2026. Helicone shipped for three years and processed 14.2 trillion tokens before entering maintenance mode. Our track record is short; our status page and monthly benchmark reports are where it accrues.
  • We do not replace request-level observability. If full-body logs, sessions, and evals are the job, we are the wrong tool on our own.
Migration shape

Start with one real request.

Cloud-gateway users point the OpenAI-compatible base URL at TrustedRouter, swap the API key, and remap model ids; classic passthrough users change per-provider base URLs and drop the Helicone-Auth and Helicone-* headers; async-logging users have nothing in the request path to change. What does not carry over: request logs, dashboards, HQL queries, prompt versions, and evals have no portable format, and TrustedRouter keeps metadata only, so export what you need before your plan's retention window (7 days to 3 months on non-Enterprise tiers) expires. Mintlify has said it will help Helicone customers migrate to another platform.

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

Is Helicone still safe to build on after the Mintlify acquisition?

It is running, and running fine for existing users: Helicone says services stay live in maintenance mode, with security updates, new model support, and bug fixes continuing (the monorepo was still receiving commits in July 2026). Feature development has stopped, the changelog's last entry is November 26, 2025, and Mintlify has offered to help customers migrate to another platform. For an existing workload that is a manageable short-term risk. For new infrastructure it is hard to justify when the vendor's own acquirer is pointing at the exit.

Does TrustedRouter replace Helicone's observability?

No. Helicone's request-level tooling (full-body logs, session traces, per-user analytics, HQL, prompt management, datasets, evals) is better than anything we offer, and that is by design: TrustedRouter's realtime path keeps no durable prompt or output content, so our analytics are metadata only: tokens, cost, latency, models, providers, status. If you need full-body logging, run a logging tool on your side of the connection, or use our opt-in Broadcast feature to send selected content to a destination your workspace controls.

Both products let me keep prompts out of logs. What is actually different?

The default and the guarantee. Helicone logs bodies by default; its Helicone-Omit-Request and Helicone-Omit-Response headers stop storage, but Helicone's own docs note the omitted content is still sent to their backend, and keeping payloads out entirely requires async logging, which removes the gateway from the request path. On TrustedRouter, realtime inference does not durably log content in the first place, and you can verify the exact gateway code handling your prompts through live attestation endpoints on GCP, AWS, and Azure (trust.trustedrouter.com). The boundary: our attestation covers the gateway build, and downstream model providers are covered by cited retention policy tiers, except the trustedrouter/e2e routes to providers running their own confidential compute with end-to-end encryption.

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.