Open-source API router · Verified August 16, 2026

TrustedRouter vs LMRouter

LMRouter is an MIT-licensed, self-hostable API router with zero inference markup and a hosted service whose repo has been dormant since September 2025 and whose model catalog froze in February 2026. TrustedRouter is an actively maintained gateway with hardware-attested execution, privacy-tier routing, and a 561-entry catalog (count includes meta routes) as of August 2026.

Compare details

8decision dimensions

6official sources

1base URL to try TR

Choose LMRouter when

Its operating model is the feature.

Choose LMRouter to self-host an MIT-licensed router under your own provider keys, or when 0% inference markup and free BYOK passthrough matter more than routing logic — and its ~8-provider catalog, frozen since February 2026, still covers the models you need.

Choose TrustedRouter when

Privacy needs evidence.

Choose TrustedRouter when you need current models with provider failover, privacy-tier routes (zdr, e2e, eu), and a hosted gateway you can verify by live attestation on three clouds rather than by reading source you cannot confirm is running.

DimensionLMRouterTrustedRouter

DeploymentMIT open source; self-host (TypeScript/Hono) or hosted at api.lmrouter.comHosted control plane, source-available (BUSL-1.1), with an attested API path

API surfaceOpenAI Chat Completions and Responses, Anthropic Messages, images, audio, embeddingsOpenAI Chat Completions and Responses plus Anthropic Messages

Model access65 entries (incl. aliases) across ~8 providers; frozen since Feb 2026; no video500+ models across ~50 providers: prepaid, BYOK, and direct

RoutingManual model selection; no documented fallback, retries, or load balancingProvider fallback plus auto, cheap, fast, free, ZDR, E2E, and EU routes with composable privacy and jurisdiction preferences

ObservabilityUsage logs for billing; no analytics or observability features documentedMetadata analytics and opt-in external broadcast

Prompt contentTransient pass-through; temporary cache for Responses state; no training on contentNo durable prompt or output logs on realtime inference; batch is opt-in encrypted retention

VerificationReadable MIT source; no certifications, trust page, DPA, or named legal entityLive gateway attestation on three clouds, bound to published source and release evidence

Billing0% markup; 4.8% + $0.35 per top-up; credits expire in 1 year; BYOK freePrepaid at provider price + 5.5% ($0.01/M floor), or BYOK

What LMRouter is

LMRouter is an MIT-licensed AI API router written in TypeScript (Hono on Node.js), created in July 2025. It ships in two forms: a self-hostable server you clone and configure with your own provider keys, and a hosted service at api.lmrouter.com with a unified key and prepaid credits. The API surface is wide for a small project: OpenAI-compatible Chat Completions, Responses, image generation and editing, audio, and embeddings, plus an Anthropic-compatible Messages endpoint. The GitHub repository carries an MIT license covering the whole router, server included.

It is effectively a one-person project. A single author wrote 146 of the repo's 148 commits, and the terms of service name "LMRouter Contributors" as the provider under California law. No company entity, address, or registration is disclosed anywhere on the site or docs.

Where LMRouter genuinely wins

Three things deserve plain credit. First, openness: LMRouter is OSI open source and self-hostable. Our gateway and control plane are source-available under BUSL-1.1, which is a weaker openness claim. If your requirement is "fork it and run it under MIT," LMRouter meets it and we do not.

Second, fees. LMRouter adds no markup to model prices ("We do not mark up the price of the models you use"), charges 4.8% + $0.35 only when you buy credits, and passes BYOK traffic through free (pricing). That is one of the cleanest fee structures in this category.

Third, a plainly written privacy policy: prompt content is "a transient pass-through," never used for training, and not stored long-term except temporary caching for stateful features such as Responses-API conversation state (privacy policy).

The state of the project, with dates

The hosted API works: on August 16, 2026, GET /openai/v1/models returned 200 with 65 model entries. Development has been dormant:

The main repo's last push was September 23, 2025 — about eleven months before this page's verification date.
The model-catalog repo's last commit was February 7, 2026. Models released after early February 2026 are not in the hosted catalog.
No GitHub release has ever been published.
The live catalog's 65 entries (including duplicate aliases such as claude-3-5-haiku-20241022 and anthropic/claude-3.5-haiku) span about 8 providers. The homepage says "hundreds of AI models"; the live list does not support that. There are no video models — the README marks video as coming soon.

None of this proves the service will vanish. It does mean an evaluator should price in continuity risk: one maintainer, no legal entity, no releases, a catalog frozen for six months, and prepaid credits that expire after one year and "may be removed from your account without prior notice."

Three differences that decide this comparison

Routing logic vs. an aggregator

LMRouter is an aggregator: you pick a model ID and it forwards the call. No fallbacks, load balancing, retries, or cost-aware routing appear anywhere in its 22-page docs tree. TrustedRouter routes: provider failover and fallback model lists are built into the routing layer, and meta ids select candidate pools — trustedrouter/auto, cheap, and fast, plus privacy-tier routes. trustedrouter/zdr restricts to providers with contractual zero-data-retention postures; trustedrouter/e2e restricts to confidential-compute providers with provider-side end-to-end encryption (currently Tinfoil and Phala); trustedrouter/eu pins EU-focused providers. Privacy and jurisdiction preferences compose per request. The full list is on /models.

Verification: readable source vs. attested running build

LMRouter's trust model is that you can read the source and, if you self-host, run it yourself. That is real. When you use the hosted service, though, nothing demonstrates that the code at api.lmrouter.com is the code on GitHub, and there are no published certifications, no trust page, and no DPA. TrustedRouter's hosted gateway publishes hardware attestation on three clouds — GCP Confidential Space, AWS Nitro Enclaves, and Azure Confidential Containers — with live endpoints anyone can use to verify that the serving build matches published source and release digests: trust.trustedrouter.com and /security. Realtime inference keeps no durable prompt or output logs; operational metadata (ids, model, tokens, latency, cost, region, key hash) is retained.

Catalog currency and operations

As of August 16, 2026, TrustedRouter's catalog lists 561 model entries (a count that includes our own meta ids) across 49 upstream providers, with exact prices on /models. Our status page publishes a 99.99% availability target with live error-budget burn rates on 5-minute to 24-hour windows, and we freeze monthly benchmark reports — July 2026 covered 274,753 samples across 44 providers and 353 models (/benchmarks/reports). LMRouter documents usage logs for billing and no analytics or observability features.

Fees, on their actual bases

The two fee models sit on different bases, so comparing the percentages directly is wrong. LMRouter charges 4.8% + $0.35 per credit purchase and 0% on inference. TrustedRouter's fee sits on inference: prepaid usage bills at the provider's token price + 5.5%, with a $0.01 per million token floor — a $1.00/M provider price becomes $1.055/M (/pricing). For top-ups of $50 or more LMRouter's total fee load is lower; below $50 the flat $0.35 makes its effective fee the higher one. LMRouter's BYOK passthrough is free; we support BYOK but do not publish a separate BYOK fee. The offsetting LMRouter caveat is the one-year credit expiry.

What we do not claim

Our attestation covers the gateway build, not downstream model providers. Providers are covered by cited policy and contract tiers; the e2e routes add provider-side confidential compute, and those are the provider's mechanisms, not our attestation.
Our gateway is not open source. It is source-available (BUSL-1.1, converting to Apache-2.0 four years after each release); the SDKs are Apache-2.0 or MIT. Every line that touches your prompt is public and inspectable.
We are young: our public repos date from late April 2026 and our benchmark history starts June 2026. LMRouter's repo is nine months older, though inactive.
We publish no SOC 2 or HIPAA certification today. Neither does LMRouter.

Migration shape

Start with one real request.

Change the base URL, key, and model IDs: LMRouter uses OpenRouter-style vendor/model slugs, and canonical vendor-prefixed ids mostly resolve on TrustedRouter; a genuinely unknown id returns MODEL_NOT_SUPPORTED rather than guessing. Anthropic-native clients point at our Messages endpoint. One real caveat: LMRouter's Responses API keeps conversation state in its temporary cache, so re-test stateful flows after the move.

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

LMRouter GitHub repo (MIT license, commit history)https://github.com/LMRouter/lmrouter
LMRouter pricing (fees, credit expiry)https://docs.lmrouter.com/pricing
LMRouter privacy policyhttps://docs.lmrouter.com/privacy
LMRouter terms of servicehttps://docs.lmrouter.com/terms
LMRouter live model list (65 entries, 2026-08-16)https://api.lmrouter.com/openai/v1/models
TrustedRouter attestation endpointshttps://trust.trustedrouter.com
TrustedRouter monthly benchmark reportsStable first-party route evidence
TrustedRouter live trust recordRelease and attestation evidence

All gateways

TrustedRouter vs OpenRouterHosted model marketplace
TrustedRouter vs Vercel AI GatewayHosted AI gateway
TrustedRouter vs LiteLLMSelf-hosted AI gateway
TrustedRouter vs Cloudflare AI GatewayHosted edge AI gateway

Questions

Is LMRouter more open than TrustedRouter?

Yes, in license terms. The entire LMRouter router is MIT-licensed and self-hostable with a documented local setup. Our gateway and control plane are source-available under BUSL-1.1 (each release converts to Apache-2.0 after four years); only our SDKs and integrations are Apache-2.0 or MIT. What we offer instead is verifiability of the hosted service: live attestation on three clouds that the running gateway matches the published source. If your requirement is to run the router yourself under an OSI license, LMRouter wins that point.

Which service costs less in fees?

The fees sit on different bases. LMRouter adds 0% to inference and charges 4.8% + $0.35 when you buy credits; unused credits expire after one year and may be removed without prior notice. TrustedRouter has no monthly plan or seat fees and bills prepaid inference at the provider's token price + 5.5% with a $0.01/M floor. For top-ups of $50 or more, LMRouter's total fee load is lower (4.8% + $0.35 works out below 5.5%); for smaller top-ups the flat $0.35 tips its effective fee above ours. Its BYOK passthrough is free, while we do not publish a separate BYOK fee. What you are paying our 5.5% for is failover, privacy-tier routing, current models, and the attested prompt path.

Can I get recent models through LMRouter?

Not through the hosted service today. Its model-catalog repo's last commit was February 7, 2026, and the live API listed 65 entries across ~8 providers on August 16, 2026 — models released after early February 2026 are absent. Self-hosters can add models themselves via LMRouter's per-model YAML configs. TrustedRouter's catalog listed 561 entries (including our meta routes) across 49 providers on the same date, with per-model prices on /models.