Skip to main content

Which model reviews your code

There is no single fixed model. Kodus runs on Bring Your Own Key (BYOK) by default on every plan, which means the model — and the provider relationship behind it — is normally yours, not ours.
We deliberately don’t pin model names in this page. The catalog moves fast; for the current recommended models see BYOK, which is generated from what the product actually offers today.
There are three paths, and which one applies changes who your data agreement is with:

Supported provider classes

Under BYOK you can point Kodus at:
  • First-party frontier providers — the major commercial labs, via their own APIs.
  • Cloud-hosted enterprise gateways — the same frontier models billed through your existing cloud account, with that cloud’s IAM, region pinning, and data-residency guarantees.
  • Aggregators — a single billing relationship in front of many models.
  • Any OpenAI-compatible or Anthropic-compatible endpoint — including specialized coding providers and models you host yourself, which never leaves your network.
The full, current list with setup instructions lives in BYOK.
A BYOK review never silently falls back to a Kodus-managed model. If your configured model fails and you haven’t set your own fallback, the review fails instead — we don’t reroute your code (or your bill) to our infrastructure.

What Kodus does with your code

  • Code is sent to the model provider only to produce that review, and only the parts needed: the diff plus the context the review requires.
  • Nothing is retained for training by Kodus.
  • Review results (suggestions, metadata) are stored so the product can show history, avoid repeating suggestions, and compute metrics.

What the provider does with your code

This is the part worth reading carefully, because under BYOK it isn’t governed by your contract with Kodus — it’s governed by yours with the provider.
  • Most commercial API tiers state they don’t train on API traffic by default. Verify this on the plan you’re actually on: consumer/free tiers often differ from paid API tiers.
  • Retention windows (for abuse monitoring, for example) are set by the provider, not by Kodus.
  • If you need a contractual guarantee (zero retention, a specific region, a signed DPA), get it from the provider directly — or use a cloud-hosted gateway or a self-hosted endpoint, where those terms are already part of your existing agreement.
Because you choose the provider, Kodus cannot make data-handling promises on their behalf. If your compliance posture depends on specific guarantees, confirm them with your provider before pointing Kodus at it.

Security measures on the Kodus side

  • Key handling: your provider key is encrypted at rest and never shown back in plain text.
  • Encryption: data encrypted in transit and at rest.
  • Controlled access: role-based access to the workspace — see Workspace Roles.
  • Self-hosting: run the whole stack in your own infrastructure if you don’t want code leaving your network at all. See Deploy Kodus.

Self-hosted telemetry

If you run Kodus self-hosted, each instance sends one anonymous daily heartbeat to telemetry.kodus.io with aggregated counters and runtime metadata — never code, identities, or anything that could trace back to your users. You can inspect the exact payload with pnpm run telemetry:preview or disable it with KODUS_TELEMETRY_DISABLED=true. Full schema, retention policy, and source code links are documented in Anonymous Telemetry. Questions? Contact our support team or join Discord.