AI & egress integrations

These variables configure Kutrion’s outbound integrations: the AI assistant and transactional email. AI is privacy-first by design and it is also off by default: AI_PROVIDER ships empty, so a fresh install has no assistance at all until an operator enables it. The only provider these variables can select is a self-hosted Ollama endpoint — there is no value here that selects a cloud model. On a self-hosted install that endpoint is one you run, so inference stays on your own network. The cloud LLM is a separate path, reached only when a cloud key is configured, the organization has opted in and its plan carries the cloud-AI entitlement. Mail is optional; with no SMTP configured the mailer is a no-op and invite links are shown inline instead of emailed. For the design and guarantees — including what the hosted service does differently — see AI privacy model.

AI_PROVIDER=ollama
OLLAMA_URL=http://192.0.2.10:11434
AI_MODEL=llama3
SMTP_HOST=smtp.example.com
SMTP_PORT=465
SMTP_SECURE=true
[email protected]
SMTP_PASS=replace-with-smtp-password
SMTP_FROM="Kutrion <[email protected]>"

AI

Variable	Type	Default	Description
AI_PROVIDER	string	(empty)	LLM provider: empty (no AI at all) or ollama (self-hosted). No cloud value exists.
OLLAMA_URL	string	(empty)	Self-hosted Ollama endpoint; required when AI_PROVIDER=ollama.
AI_MODEL	string	llama3	Local model name at Ollama.
ANTHROPIC_API_KEY	string	(empty)	Cloud LLM key (Claude). A key alone does not enable cloud egress — the organization must also opt in and hold the cloud-AI entitlement. Empty disables cloud.
CLOUD_AI_MODEL	string	claude-haiku-4-5-20251001	Cloud model id for the opt-in cloud fallback.
AI_RATE_MAX	number	60	Max AI requests per window.
AI_RATE_WINDOW_MS	number	60000	AI rate-limit window.

AI_PROVIDER is empty as shipped, which is no AI at all. Set it to ollama and point OLLAMA_URL at an endpoint you run, and all inference happens on that host — terminal output never leaves your network in this configuration, and it is this configuration, not a product default, that keeps it there. These variables belong to whoever operates the deployment, so on the hosted service they are Kutrion’s and not yours; the tenant-side equivalent is the per-organization AI setting described in AI privacy model. The cloud path needs three things together: ANTHROPIC_API_KEY empty disables cloud entirely, and even when a key is present the cloud model is invoked only for an organization that has opted in and whose plan carries the cloud-AI entitlement. Secrets are stripped from anything that does go out, but that step is a filter rather than a gate — see AI privacy model for what it removes and what it sends. OLLAMA_URL is operator configuration, so the server calls it directly without passing it through the private-network egress guard: a LAN or localhost Ollama works without EGRESS_ALLOW_PRIVATE. The guard does apply to an Ollama endpoint an organization sets in its own AI settings, which is checked when it is saved and again on every call; if your organizations point those at private addresses, set EGRESS_ALLOW_PRIVATE=1 (see Core & database configuration). For setup, see Self-hosted AI setup.

Mail (transactional)

Variable	Type	Default	Description
SMTP_HOST	string	(empty)	SMTP server; empty makes the mailer a no-op (invite links shown inline).
SMTP_PORT	number	465	SMTP port.
SMTP_SECURE	boolean	true	Use TLS/SSL.
SMTP_USER	string	(empty)	SMTP username.
SMTP_PASS	string	(empty)	SMTP password.
SMTP_FROM	string	(example sender)	Envelope sender and display name.

Transactional email is used for invitations and verification. If SMTP_HOST is empty the mailer is a no-op and the relevant links (for example an invite link) are surfaced inline in the UI instead of being emailed — useful for an air-gapped install with no mail server. To send mail, set SMTP_HOST, credentials, and a SMTP_FROM address. Use SMTP_PORT=465 with SMTP_SECURE=true for implicit TLS, or adjust both for your provider’s STARTTLS port.