Skip to main content
Coding agents work out of the box with UncensoredGPT — every model in the catalog supports tool calling, and the API speaks both the OpenAI and Anthropic formats. Setup takes about a minute.

The universal recipe

Any agent that accepts a custom OpenAI-compatible provider needs exactly three things:
  1. Base URL — https://api.uncensoredgpt.ai/v1
  2. API key — create one in the Console and export it as an environment variable:
  3. Model ID — the exact IDs returned by List models, e.g. uncensoredgpt/qwen3.8-27b-uncensored. Never invent or shorten them.
That’s it. If your agent isn’t listed below, point it at the base URL with your key and pick a model — it will just work.

Guides per agent

We recommend starting with OpenCode — it’s free and open source, connects to UncensoredGPT with a single config file, and supports all ten models out of the box.

Available models

All uncensoredgpt/ models run on our own infrastructure — no filters, no logs, and your code never leaves our servers.
The free model is for testing our servers — it is not uncensored and reliability is not guaranteed. For agent work, pick one of the uncensored models. The full catalog — including third-party and free models — is on the Model catalog page.

Good to know

  • Tool calling is required by agents and supported by every model above.
  • Streaming is supported and enabled by agents automatically.
  • Per-key limits (requests/minute, tokens/minute) apply to agent usage too — see Rate limits. Agents retry on 429, but raise your limits if you run long sessions.
  • Costs are per token and visible per request in the Console’s Usage page. Long agent sessions consume a lot of context — the 1M-context models are a good fit.