Skip to content
Product9 min readRepath Khan

LLM Gateway BYOK on Critique: Same $8 Harness, Broader Model Surface

Critique now supports LLM Gateway alongside OpenRouter and CrofAI. Save one key in Settings, keep the same review and chat flow, and let LLM Gateway bill the model calls directly.

The original BYOK story on Critique was simple. If you wanted the product system but not Critique carrying every model invoice, you could bring OpenRouter. Then we added CrofAI for teams that wanted a more OSS-first billing path.

That still left a gap. Some teams want a broader OpenAI-compatible gateway without changing how Critique review policies, sandbox runs, and chat sessions already behave. LLM Gateway is that lane.

$0
Monthly BYOK harness, Critique orchestration only
0
Models listed by LLM Gateway in provider docs on 29 June 2026
0
Providers listed by LLM Gateway in those docs
0
Review-capable models Critique currently exposes in product catalog

Because BYOK is not one market. Some teams want frontier model routing and already centralize budgets on OpenRouter. Some want direct OSS-host billing on Crof. Others want a wider multi-provider OpenAI-compatible gateway while keeping the same product harness for sandboxes, OpenCode review turns, GitHub checks, retrieval, and usage ledgers.

Critique should not force those three intentions through one provider if the rest of the product can stay constant. The point of the harness is to keep the review system stable while the upstream billing lane stays flexible.

How the three BYOK lanes now differ on Critique

The product workflow stays the same. The upstream bill, catalog shape, and best-fit use case are what move.

LaneUpstream billBest fit
OpenRouter BYOKOpenRouter accountTeams already standardizing frontier and mixed-provider spend on OpenRouter.
CrofAI BYOKCrofAI accountTeams optimizing around OSS-hosted model economics and Crof-native routes.
LLM Gateway BYOKLLM Gateway accountTeams wanting a broad OpenAI-compatible multi-provider gateway without changing Critique review flow.

All three use the same Critique harness. The harness fee covers orchestration, not model tokens.

Critique with managed billing
Pick model in CritiqueCritique review or chat flowCritique-managed inference bill
Critique with LLM Gateway BYOK
Pick same model in CritiqueCritique review or chat flowapi.llmgateway.io/v1Your LLM Gateway bill

The important part is what does not change. You do not learn a new review surface. You do not rewrite policy just because the bill moves. Critique keeps the same model picker, the same review and chat orchestration, the same sandbox path, and the same usage ledger.

Under the hood, Critique strips OpenRouter-specific prefixes and route suffixes from request model ids, sends the normalized id to https://api.llmgateway.io/v1, and omits OpenRouter-only headers. That is why the feature can behave like a gateway swap instead of a second product.

This is a billing and routing feature, not a promise that every model listed by LLM Gateway suddenly appears in every Critique dropdown. Today Critique exposes 35 review-capable models in its own catalog. LLM Gateway publishes a much larger provider surface than that.

That distinction matters. The gateway is broader than the current product catalog. The product catalog is broader than what some provider-specific lanes expose. We are not pretending those are the same thing.

It also does not replace BYOA. Saving an LLM Gateway key changes which account pays for review and chat inference. It does not replace Cursor, Codex, Claude Code, or other external fix-agent execution lanes.

LLM Gateway BYOK checklist
  1. 1
    Subscribe to the BYOK harness
    Use the $8/month Critique harness if you want Critique to keep running orchestration while your provider bills the model calls.
  2. 2
    Create an LLM Gateway API key
    Generate the key in your LLM Gateway account and apply whatever spend limits or provider policy you want there.
  3. 3
    Save the key in Critique
    Open Settings → Agents → Bring your own LLM Gateway key, paste the key, and save it encrypted. The browser only keeps a hint after save.
  4. 4
    Run one chat turn and one review
    Verify that Usage shows the model calls as externally billed where applicable and that the workflow feels identical from the operator side.
  5. 5
    Know the fallback order
    If you also keep CrofAI or OpenRouter keys saved, LLM Gateway wins until you remove it. No review-policy rewrite is required.
No. The Critique harness fee is the same $8/month lane. The difference is which upstream provider bills the model tokens.
Those lanes still work. LLM Gateway is additive. If you save all three keys, Critique resolves the active provider in this order: LLM Gateway, then CrofAI, then OpenRouter.
No. Critique currently exposes 35 review-capable models in the product catalog. The gateway surface behind that catalog is wider than the current in-product list.
BYOK changes who pays for inference on review, Remedy, Builder, and chat model calls. BYOA changes which external agent account runs fix work such as Cursor, Codex, or Claude Code.

Add the key and keep the workflow

If your team wants Critique to stay the review system while LLM Gateway carries the model invoice, the path is now live in Settings.

Open Settings →