AI Features and BYOK

What Navis Docs AI does

Navis Docs uses AI to help teams find answers in their own procedures and to speed up procedure creation. The AI chat inside each team answers questions using that team's published procedure content.

Behind the scenes, procedures are split into chunks and embedded with OpenAI text-embedding-3-small. When a user asks a question, Navis Docs performs semantic search over those chunks and sends the most relevant context to the chat model. This retrieval-augmented generation workflow is commonly called RAG.

Provider roles

Navis Docs uses different AI providers for different jobs. This keeps search and answer generation separate.

• OpenAI: used for procedure embeddings with text-embedding-3-small
• Anthropic: used for AI chat and AI-assisted procedure import with claude-haiku

Embeddings make procedure content searchable by meaning. Anthropic powers the conversational responses and import assistance that turn uploaded or pasted material into usable procedure drafts.

Bring your own key (self-hosted)

On self-hosted deployments, organization owners and admins can provide AI API keys in organization settings. This is the bring-your-own-key, or BYOK, model. It lets each organization control the keys used for its AI features instead of relying only on environment variables.

Stored keys are encrypted at rest using AI_KEY_ENCRYPTION_SECRET. Navis Docs can check whether a key exists, but it does not expose the raw value after it has been saved.

Which key enables which feature

The required key depends on the feature you want to use.

• Anthropic key: enables AI chat and AI-assisted procedure import
• OpenAI key: can be stored in organization settings, but embeddings currently use the server environment key

If AI chat or procedure import is unavailable, check that the organization has an Anthropic key configured or that the self-hosted environment provides one.

Cloud and self-hosted deployments

Navis Docs cloud does not offer AI features. There is no AI chat, AI-assisted procedure import, or procedure chunk embeddings on cloud. Published procedures are not RAG-indexed on cloud.

AI features are available on self-hosted deployments only. Operators set platform keys in the environment (ANTHROPIC_API_KEY for chat and import, OPENAI_API_KEY for embeddings), or organizations configure keys in Settings (BYOK). Both Anthropic and OpenAI keys are required for chat and semantic search. When the server environment provides keys, users do not need to configure BYOK inside the organization. For deployment setup, see Getting Started with Self Hosting.

Managing saved keys

Owners and admins can remove AI keys at any time from organization settings. Removing a key disables the features that depend on it until a valid key is added again or supplied by the self-hosted environment.

The key status shown in the app comes from getAiKeyStatus. It confirms whether a key is present and usable for configuration checks, but it never returns the secret key value to the browser.

Suggested setup workflow (self-hosted)

1. Decide how keys will be supplied

Self-hosters can choose between organization-provided keys in Settings and environment-level keys in .env, depending on how you want to manage secrets. Set NEXT_PUBLIC_DEPLOY_MODE=self-hosted and provide both ANTHROPIC_API_KEY and OPENAI_API_KEY at the platform level, or configure them per organization in Settings.

2. Add the Anthropic key

Add an Anthropic key first if you want AI chat or AI-assisted procedure import. These are the user-facing AI features most teams enable.

3. Confirm AI availability

Open a team with published procedures and try the AI chat. Good answers depend on clear, published procedure content because the chat searches the team's procedure chunks before responding.

4. Rotate or remove keys when needed

If a key is rotated at the provider, update it in Navis Docs. If your organization stops using AI features, remove the saved key from organization settings.