Skip to content
9 min readCritique

v3.6.0: Bring Your CrofAI Key — Same $8 Harness, Direct OSS Billing

Critique BYOK now routes chat and sandbox PR review through CrofAI (nahcrof) or OpenRouter. Pay $8/month for the harness; pay Crof or OpenRouter for the tokens.

Crof + OpenRouter BYOK

$8 harness · models at your cost

critique.sh

When we shipped the $8/month BYOK harness in May, the story was simple: Critique runs the review system; OpenRouter bills the models. That lane still exists — and it is still the right choice if you want frontier GPT or Claude routes, transcription, or a single OpenRouter budget across tools.

A growing slice of teams does not need OpenRouter in the middle for every pass. They want direct OSS hosting: Kimi K2.6, DeepSeek V4, GLM 5.1, MiMo, and Crof-native models like Greg 1 — billed by nahcrof/Crof at provider-native rates. v3.6.0 wires that path into the same product harness you already trust for sandbox review and chat.

What you pay Critique vs what you pay the model host

Bundled Solo / Pro / Team plans are unchanged. BYOK is a separate lane for teams that want direct provider invoices.

Line itemCritiqueYour provider
Harness (orchestration)$8/mo USD · €8/mo EUR$8/mo USD · €8/mo EUR
Model tokens (chat + sandbox review)OpenRouter or CrofAI account
Typical upstreamE2B, OpenCode, GitHub App, retrievalSame product stackopenrouter.ai or crof.ai/v1
Best forPredictable harness feeTeams with existing OpenRouter spendTeams optimizing OSS model cost on Crof

Student/OSS ($5/mo) and bundled credit tiers (Solo $19, Pro $49, Team $149) are separate products. BYOK is for direct model billing with Critique orchestration.

OpenRouter BYOK
Policy picks moonshotai/kimi-k2.6Critique harnessopenrouter.ai/api/v1Your OpenRouter bill
Crof BYOK
Same policy id in UICritique harnesscrof.ai/v1 → kimi-k2.6 slugYour Crof bill

Critique’s model picker stays OpenRouter-shaped because policy, credits, and catalog docs were built around those ids. At request time we map known ids to Crof slugs — for example moonshotai/kimi-k2.6 becomes kimi-k2.6. That is why the in-app list can look narrower than Crof’s live catalog even though BYOK unlocks more OSS models on the provider side.

Frontier OpenAI and Anthropic models in Critique policy are not available on Crof’s OpenAI-compatible endpoint. If your review policy depends on those lanes, keep an OpenRouter key (or use bundled Team credits). Crof BYOK shines when your lead and specialist stack is OSS-first and you want nahcrof’s direct hosting economics.

If you save both Crof and OpenRouter keys, Crof takes precedence. That matches the typical intent: you added Crof because you want that billing path. Remove the Crof key in Settings and Critique falls back to OpenRouter BYOK automatically — no policy rewrite required.

Crof BYOK checklist
  1. 1
    Subscribe to the harness plan
    On Pricing, choose the $8/month BYOK harness (€8/mo in Europe). That covers Critique orchestration — not model tokens.
  2. 2
    Create a Crof API key
    Sign in at ai.nahcrof.com, create a key, and set spend limits you are comfortable with.
  3. 3
    Save the key in Critique
    Settings → Bring your own CrofAI key → Save encrypted key. Keys never return to the browser after save.
  4. 4
    Optional: keep OpenRouter for frontier
    Save an OpenRouter key as fallback for GPT/Claude lanes. Remember Crof wins when both exist — remove Crof temporarily if you need a pure OpenRouter run.
  5. 5
    Verify Usage
    Run one chat turn and one sandbox PR review. Usage should show externally billed model calls with zero Critique credits on the BYOK path.
No. The Critique harness is the same $8/month ($8 USD, €8/mo EUR, and localized equivalents on the pricing page). Only the upstream model invoice changes — Crof bills Crof; OpenRouter bills OpenRouter.
Yes, on different paths. BYOK model calls on chat and sandbox review skip Critique credits when your key is active. Indexing, Remedy on managed models, and other non-BYOK surfaces may still draw from your monthly pool.
Policy and the UI use OpenRouter catalog ids for consistency. Crof exposes additional OSS models via GET https://crof.ai/v1/models. Mapped ids work automatically; frontier GPT/Claude picks require OpenRouter BYOK or bundled Team credits.
BYOK pays for inference on review and chat. BYOA queues fix agents on your Cursor, Anthropic, or OpenAI subscriptions. Teams often pair Crof or OpenRouter BYOK for review depth with BYOA for execution.

Wire your key and run a review

Open Settings to add Crof or OpenRouter, confirm the harness plan on Pricing, and read the full BYOK guide in docs.

Open Settings →