05 / REFERENCE

Environment variables

Configure NConnect with environment variables: API keys, base URL, reasoning effort, fallback model, daemon metering, auto-update and telemetry.

Keys resolve in this order: the --api-key flag, then the key saved by nconnect configure, then the environment.

Variable	Effect
`NEBIUS_API_KEY`	Nebius key, used when none is saved by `configure` or passed with `--api-key`.
`TAVILY_API_KEY`	Enables web search (or set it via `configure`).
`NEBIUS_BASE_URL`	Override the API base. Default `https://api.tokenfactory.nebius.com/v1`.
`NCONNECT_REASONING_EFFORT`	`none` \| `low` \| `medium` \| `high` \| `max`. Default `none` for speed; raise it for harder tasks.
`NCONNECT_FALLBACK_MODEL`	Model to fail over to when the target returns no response headers. Default `moonshotai/Kimi-K2.6`; set `off` to disable.
`NCONNECT_METER=1`	Route the spawned harnesses through the daemon so they get cost metering, model fallback and retries. Off by default.
`NCONNECT_REASONING_HISTORY`	`full` (default) \| `interleaved` \| `off`. How much prior reasoning is replayed each turn. `off` is cheapest on long sessions.
`NCONNECT_CACHE_READ_RATIO`	Price of a cached input token as a fraction of the input price. Default `1`, since Nebius publishes no cached rate, so the total is an upper bound.
`NCONNECT_CODEX_MEMORY_MODEL`	Model that summarizes Codex traces for durable memory. Defaults to MiniMax M3.
`NCONNECT_DISABLE_AUTOUPDATE=1`	Stop the installed binary from self-updating.
`NCONNECT_TELEMETRY_URL`	Opt in to telemetry by pointing at your own collector. Off by default.