Custom LLM

Bring your ownOpenAI-compatible model

On paid plans, route CodeCritic reviews through your HTTPS OpenAI-compatible endpoint. You keep provider choice and contracts; CodeCritic still orchestrates reviews, quotas, and the same finding UX across paste, API, and GitHub.

OpenAI-compatible HTTPSpaid plansSettings → custom LLMsame finding UXyou pay provider tokens

Policy-friendly routing

When procurement requires a specific vendor or private gateway, point CodeCritic at that base URL and model instead of platform defaults.

Orchestration stays with us

Diff packing, finding structure, Policy findings, Trust Focus, and PR comments remain CodeCritic workflows - your endpoint supplies completions.

Clear cost split

Platform review quota still applies. With custom LLM enabled, token spend for inference is billed by your provider.

Workflow

How teams wire it

  1. 1

    Upgrade to a paid plan

    Custom LLM is not included on Free. Compare tiers on Pricing.

  2. 2

    Enable in Settings

    Enter base URL, model, and API key. Save, then run a small paste review to verify.

  3. 3

    Watch the review UI

    Results still show scores, issues, Policy findings, and Applied policy - same as platform models.

  4. 4

    Disable when done testing

    Remove custom LLM config to fall back to CodeCritic platform models.

Overview

Not the same page as the REST API

The API landing is about calling CodeCritic to create reviews from your systems. This page is about which model CodeCritic calls on your behalf when a review runs.

Free plans use CodeCritic platform models only. Custom LLM configuration lives under Dashboard → Settings on paid tiers.

Operational constraints

Base URL must be public HTTPS. You provide model id and API key (stored via Secret Storage / encrypted settings). Clear custom LLM to return to platform models.

If platform quota is exhausted while custom LLM is on, reviews can still use your provider - you pay those tokens. Watch both CodeCritic usage and vendor billing.

Checklist

  • Public HTTPS only Private VPC hostnames that CodeCritic cannot reach will fail the review job.
  • Compatible chat completions Endpoint should speak OpenAI-compatible request/response shapes CodeCritic already targets.
  • Secrets Never paste provider keys into PR comments or client-side logs.

FAQ

Common questions

No. Free uses CodeCritic platform models. Custom OpenAI-compatible routing is a paid-tier option.

Route reviews through your gateway

Upgrade when ready, paste your OpenAI-compatible base URL in Settings, and verify with one paste review.

Integrations