5. Configuration

Config home directory

Rayu resolves a single config home directory used for settings, providers, sessions, skills, projects, etc. Resolution order:

RAYU_CONFIG_DIR — explicit override.
~/.rayu — if it exists (Rayu's own config).
~/.rayu — default for fresh installs.

Force a different directory with the env var:

RAYU_CONFIG_DIR=/tmp/rayu-test rayu  # isolated/throwaway config

Files & locations

Within the config home (e.g. ~/.rayu/):

Path	Purpose	Notes
`providers.json`	Rayu providers: id, kind, apiKey, baseURL, default/fetched models, context overrides	mode `0600` (secrets)
`settings.json`	User settings, incl. the selected `model`
`diagnostics.jsonl`	Recorded bugs/issues/vulnerabilities	append-only JSONL
`projects/`	Per-project session transcripts
`skills/`, `agents/`, …	Skills/agents/etc.

When RAYU_CONFIG_DIR is set, all config files live inside that directory instead of the default ~/.rayu.

`providers.json` schema

{
  "activeProvider": "nvidia",
  "providers": [
    {
      "id": "nvidia",
      "kind": "openai-compatible",
      "apiKey": "nvapi-xxxxx",
      "baseURL": "https://integrate.api.nvidia.com/v1",
      "defaultModel": "meta/llama-3.3-70b-instruct",
      "smallFastModel": "meta/llama-3.3-70b-instruct",
      "models": ["my/custom-model"],
      "fetchedModels": ["...catalog from /v1/models..."],
      "contextWindow": 131072,
      "modelContextWindows": { "deepseek-ai/deepseek-v4-flash": 1000000 }
    },
    { "id": "anthropic", "kind": "anthropic", "apiKey": "sk-ant-xxxxx" }
  ]
}

Field	Meaning
`activeProvider`	id of the provider currently in use
`kind`	`anthropic` or `openai-compatible`
`apiKey`	provider API key (secret)
`baseURL`	endpoint base (openai-compatible)
`defaultModel`	model used until you switch / fallback
`smallFastModel`	model for cheap requests (titles, etc.); defaults to `defaultModel`
`models`	user-pinned model ids (shown in `/model`)
`fetchedModels`	cached catalog from `GET {baseURL}/models`
`contextWindow`	provider-wide context default (tokens)
`modelContextWindows`	per-model context overrides (tokens)

You can edit this file by hand; restart Rayu to pick up changes.

Environment variables

Config & providers

Variable	Effect
`RAYU_CONFIG_DIR`	Override config home dir
`RAYU_OPENAI_COMPATIBLE=1`	Force the OpenAI-compatible client path
`RAYU_OPENAI_BASE_URL`	Base URL for the OpenAI-compatible endpoint
`RAYU_OPENAI_API_KEY`	API key for the OpenAI-compatible endpoint
`ANTHROPIC_API_KEY`	Anthropic API key
`NVIDIA_API_KEY`, `DEEPSEEK_API_KEY`, `KIMI_FOR_CODE_API_KEY`, `DOUBLE_WORD_API_KEY`, `OPENAI_API_KEY`, `OPENROUTER_API_KEY`	Auto-imported into `providers.json` on startup

Models & context

Variable	Effect
`RAYU_CONTEXT_TOKENS`	Force the context window (tokens) for the active model
`ANTHROPIC_MODEL`	Default model id (Anthropic-style precedence)
`ANTHROPIC_SMALL_FAST_MODEL`	Override the small/fast model

Diagnostics & privacy

Variable	Effect
`RAYU_DIAGNOSTICS=1`	Also echo diagnostics to stderr
`RAYU_DIAGNOSTICS_NO_FILE=1`	Don't persist diagnostics to disk
`RAYU_TELEMETRY=1`	Opt back into telemetry (off by default)
`DISABLE_TELEMETRY`	Force telemetry off (`no-telemetry`)

See Diagnostics & Privacy for the privacy model.

Project settings & memory

Rayu reads project files when present:

RAYU.md, .rayu/RAYU.md, .rayu/rules/*.md — project memory/instructions.
.rayu/settings.json, .rayu/settings.local.json — project/local settings.
.mcp.json — project MCP servers.

Next: CLI Reference →