Model providers

OpenKyrozen separates provider selection, credentials, model defaults, and task-complexity routing. The provider registry in openkyrozen/providers/registry.py is the source for selectable names and repository defaults. A model shown here is OpenKyrozen's configured default; accounts, regions, and provider catalogs can change independently. Override a model when the selected provider does not expose its default in your account.

Current registry defaults

Confirm names, default values, credential lookup, and optional extras against the current provider registry, pyproject.toml, and configuration before changing them. Providers that use deployment names, ambient credentials, or regional catalogs intentionally leave model defaults blank in the registry.

Configure credentials and defaults

Set the provider and its credential in the shell that starts OpenKyrozen:

export KYROZEN_PROVIDER=deepseek
export DEEPSEEK_API_KEY=replace-with-your-key
export KYROZEN_MODEL_SIMPLE=deepseek-flash
export KYROZEN_MODEL_COMPLEX=deepseek-v4-pro
kyrozen

The interactive provider and /api_key flows can save credentials through OpenKyrozen's encrypted user configuration. Treat that file as private data. Environment settings are useful for servers and automation; do not place keys in agent.yaml, project files, command-line arguments, shell history, or issue reports.

OpenAI-compatible services can use KYROZEN_BASE_URL where the selected integration supports it. Ollama normally listens at http://localhost:11434/v1; override its endpoint and model names if your server differs. Azure OpenAI, Bedrock, and Vertex need account-specific resource/deployment, region, or identity configuration; installing the optional Python extra alone does not provision cloud access.

Use /custom-provider create to add a named OpenAI-compatible endpoint with its optional key, endpoint URL, simple and complex model IDs, and context limit. /custom-provider list, edit <name>, use <name>, and remove <name> manage profiles. The TUI provider picker opens the same guided setup with masked key entry. Profiles are stored in ~/.kyrozen_config.json; API keys are encrypted. A sub-agent role can select a profile with provider: custom and custom_profile: <name> in agent.yaml.

Model routing and fallback

The normal provider setting has separate simple and complex model slots. The agent chooses between them using its task routing policy. /model <name> pins the main agent to one model; /model auto restores complexity routing. Sub-agent model assignments remain independent. Ollama never selects an installed model automatically: setup lists installed tags as suggestions and requires a chosen tag. KYROZEN_CONTEXT_WINDOW_TOKENS supplies a context size when an otherwise unknown custom model cannot be detected; use the actual provider limit, including the space needed for output.

The registry marks a subset of providers for automatic selection and records preferred fallback providers. A fallback is not guaranteed: it still needs credentials, an available compatible model, and a supported transport. Alternate sub-agent providers use their own configured environment credentials and do not inherit a generic main-provider key. See sub-agents before using cross-provider assignments.

Optional dependencies

pip install -e '.[web]' installs FastAPI and Uvicorn. Provider extras are .[claude], .[gemini], .[perplexity], .[bedrock], and .[vertex]; .[cloud] groups the cloud integrations. .[browser] installs Playwright; browser execution also needs a browser installation. .[all] installs the supported optional Python integrations. The official installer has its own pinned set of dependencies; see installation.

Costs and changing provider catalogs

The cost ledger attributes usage to the provider and model returned by the integration, including streaming usage when provided. Displayed costs can be estimates when a provider does not report token usage or published pricing is missing. Do not use a README model label or price as evidence that your account can access that model. See usage and cost for price provenance and interpretation.