Kimi Code | OcHub

Manage Kimi Code providers, models, limits, capabilities, and native TOML configuration.

OcHub manages Kimi Code as a current-connection tool. Each OcHub connection is a complete provider-and-model profile. Switching writes the selected profile to ~/.kimi-code/config.toml, updates default_provider and default_model, and keeps unrelated top-level Kimi Code settings intact.

Add a connection

Open Kimi Code in the OcHub sidebar and select Add connection.
Enter a stable Provider ID. Models reference this value through their provider field.
Select the protocol used by the upstream: OpenAI Chat Completions, OpenAI Responses, Anthropic Messages, Kimi, Google GenAI, or Vertex AI.
Enter the Base URL and API key required by that provider.
Add at least one model. Each row needs a unique alias, upstream model ID, and positive context-window size. Output limit, display name, and capabilities are optional.
Set Default model alias to one of the aliases in the model table.
Review the config.toml preview, save, and select Switch.

Capabilities are comma-separated identifiers such as thinking, tool_use, and image_in. Custom headers are sent with provider requests. Credential environment mappings are written under the provider’s env table; they do not modify the shell environment.

Existing configuration

When Kimi Code already has a valid active provider, OcHub can discover it as a connection. The imported profile contains that provider and the models that reference it. Comments and unrelated sections are preserved when OcHub writes the selected profile, while stale provider/model entries from the previously managed profile are replaced.

Back up config.toml before the first switch if it contains hand-written provider layouts that you want to retain. Kimi Code support does not currently include MCP or skill synchronization.

Verify

After switching, restart Kimi Code, confirm the expected model is selected, and send a short request. If it still uses an old endpoint, verify that OcHub and Kimi Code resolve the same home directory and that default_provider and default_model point to entries present in the generated file.