Managed setup defaults

Organizations can preload non-secret provider, model, and MCP choices into guided setup. Existing user configuration takes precedence, and users review and validate every connection before saving.

Never include secrets

Do not put API keys, tokens, or passwords in this payload. Environment and named keyring references are safe because they only identify where setup should find a secret; a template without a source prompts for a new keyring value.

The payload accepts inference_providers (provider objects with an additional name), llm_model, emb_model, and mcp_servers (configurations keyed by server name).

{"inference_providers":[{"name":"company","base_url":"https://models.example.com/v1"}],"llm_model":{"model":"gpt-5.4","model_provider":"openai","inference_provider":"company"},"mcp_servers":{"company_tools":{"transport":"streamable_http","url":"https://tools.example.com/mcp","headers":{"Authorization":{"template":"Bearer %s"}}}}}

Download before passing --config/-Config: Linux/macOS: curl -LsSf https://lanl.github.io/ursa/latest/install.sh -o install.sh && sh install.sh --config defaults.json --version 1.2.3 --python 3.12; Windows: irm https://lanl.github.io/ursa/latest/install.ps1 -OutFile install.ps1; .\install.ps1 -Config defaults.json -Version 1.2.3 -Python 3.12.

Omit version to install the latest ursa-ai; use --version/-Version for a PyPI release or --branch/-Branch for a LANL URSA branch. Targets are passed to uv without validation. --python/-Python defaults to 3.13. Setup defaults never select a package source or version; legacy python_version, extras, and extra_packages fields remain accepted but do not affect installation.