Skip to main content

​
How Chutes works

Chutes AI is a decentralized serverless compute platform for open-source models. It exposes an OpenAI-compatible inference endpoint and offers subscription plans that bundle API usage up to a cap expressed as a multiple of the equivalent pay-per-token value — similar in structure to the Z.AI GLM Coding Plan, but covering the full open-source catalog (DeepSeek, Llama, Qwen, MiniMax, Kimi, and many more). Kodus talks to Chutes through the same OpenAI-compatible adapter it uses for everything else, so there are no code changes — just BYOK credentials.

​
Plans at a glance

Pricing and quota rules change. Always confirm at chutes.ai/pricing before choosing a tier.
Since early 2026, every Chutes subscription includes a usage allowance equal to 5× the equivalent pay-as-you-go value of the tier, calculated from each model’s per-million-token price. Representative tiers (confirm current numbers on the pricing page):
  • The 5× cap resets monthly and is computed against the same per-token prices you’d pay pay-as-you-go.
  • Some models require Standard or higher — the base tier does not carry frontier coding models.
  • Chutes marks some models with a -TEE suffix indicating trusted-execution-environment (confidential compute) variants.
Chutes uses HuggingFace-style org/model identifiers, sometimes with a -TEE suffix for the confidential-compute variant: See the live list and current pricing at llm.chutes.ai/v1/models.

​
Creating an API Key

A Chutes account with an active subscription (or pay-as-you-go balance) is required.
  1. Go to chutes.ai and create an account.
  2. Subscribe to a tier at chutes.ai/pricing, or enable pay-as-you-go if you prefer.
  3. Open the developer console and create an API key. Copy it immediately.

​
Configure Chutes in Kodus

  1. In the sidebar, open AI providers (app.kodus.io/byok).
  2. Chutes has no dedicated tile — on the Providers tab, pick OpenAI-compatible under Custom. Which model runs by default and which is the fallback is set afterwards on the Routing tab.
  3. Fill the form:
  4. Click Test & save. Kodus probes the endpoint and persists the config on success.
The 5× cap is computed from per-token prices. Expensive frontier models burn through the cap faster than small ones — if you want to maximize reviews per dollar, pair Chutes with a cheaper model (Llama, Qwen smaller variants) for routine PRs and save the frontier models for complex reviews via a per-agent override on the Routing tab or a per-repository Code review model (Code review → General).
Because Chutes runs on decentralized compute, cold-start and tail latency vary more than on dedicated providers. Set an OpenAI or Anthropic model as Fallback (optional) on the Routing tab so Kodus can fail over when a node is slow or the monthly cap is hit.

​
Option 2 — Self-hosted (environment variables)

If you run Kodus in Fixed Mode (single global provider, no per-org BYOK), configure Chutes in the .env of your API + worker containers:
This path is only needed for self-hosted Kodus installs that deliberately disable BYOK. If BYOK is enabled on your self-hosted instance, prefer Option 1 — the UI-based flow is the same as on Cloud.
Restart the API and worker containers after editing .env, then verify:
For the full self-hosted setup (domains, security keys, database, webhooks, reverse proxy), follow the generic VM deployment guide and only swap the LLM block for the one above.

​
When to pick Chutes

  • You want the broadest open-source catalog at a subscription price — frontier DeepSeek / MiniMax / Qwen at a flat fee with predictable caps.
  • You care about confidential compute — Chutes offers -TEE variants that run inside trusted execution environments, useful if your compliance posture requires it.
  • You’re running at low-to-mid volume and fit within the 5× PAYG cap of a cheap tier.
Pick Synthetic instead if you want a simpler flat subscription with no per-model cap math. Pick Z.AI if your preferred model is the GLM family specifically.

​
Troubleshooting

  • Frontier models (DeepSeek V3, MiniMax M2.1, some Qwen variants) are gated to the Standard tier and above since Feb 2026.
  • Either upgrade, or pick a model available on your current tier (smaller Llama or Qwen variants).
  • Check current usage in the Chutes dashboard.
  • Switch temporarily to a cheaper model to extend the cap, or upgrade tier.
  • Set a Fallback (optional) model on the Routing tab so reviews keep running while you’re capped.
  • Confirm the key is active in the Chutes dashboard and the subscription is current.
  • Make sure there are no trailing spaces or quotes in the .env value.
  • Chutes uses the org/model format, with some variants ending in -TEE (confidential compute). Double-check exact capitalization at llm.chutes.ai/v1/models.
  • Chutes runs on decentralized compute, so tail latency is higher than on dedicated clouds.
  • For latency-sensitive reviews, prefer dedicated providers; reserve Chutes for overnight or batch review jobs, or set a fast provider as Model for all tasks and Chutes as Fallback (optional) in Routing.
  • Confirm your server can reach llm.chutes.ai.
  • Review API and worker logs for the exact upstream error.