Uncensored models × coding agents compatibility matrix
Every hosted uncensored model ranked here works in Claude Code, Codex, OpenCode, Cline, Kilo Code, Roo Code and Hermes Agent, and in Cursor's chat.
Each cell links to that agent's setup page, and local models depend on the server you run.
What each agent needs
Pick the agent first, then an API that serves its format.
| Agent | Format | Endpoint | Configured in | Switch models with |
|---|---|---|---|---|
| Claude CodeAnthropic's terminal agent | Anthropic Messages | /v1/messages | ~/.claude/settings.json | /model and /effort |
| CodexOpenAI's coding agent | OpenAI Responses | /responses | ~/.codex/config.toml | /model |
| OpenCodeOpen-source terminal agent | OpenAI Chat Completions | /chat/completions | opencode.json | /models and ctrl+t |
| CursorAI code editor | OpenAI Chat Completions | /v1/chat/completions | Cursor Settings | the model picker |
| Cline, Kilo Code, Roo CodeVS Code extensions | OpenAI Chat Completions | /chat/completions | Extension settings | the extension settings |
| Hermes AgentNous Research's agent | OpenAI Chat Completions | /chat/completions | ~/.hermes/config.yaml | /model and /reasoning |
Which APIs serve each format
All seven APIs for coding agents serve Chat Completions, and the four without Anthropic Messages or Responses work in Chat Completions agents only.
Local models in coding agents
A model on your machine needs a local server that speaks the agent's format. Ollama "supports a subset of the Anthropic Messages API" for Claude Code (Ollama docs). Codex's --oss flag runs it "against a local 'open source' provider such as Ollama or LM Studio" (Codex docs). The other agents take any Chat Completions server, which Cline calls the "OpenAI Compatible" provider (Cline docs).
See local models for which ones fit the GPU you have.
Notes per model
- Kimi K3.1 Pro reads video only on
/chat/completions, so not in Claude Code. - GLM-5.3 takes text only.