AI Mode and Assistant
AI Mode is a role-aware assistant for inspecting and operating SDM through the catalog available to the signed-in user.
Where to find it: Open AI Mode from the panel header; root configures it from the AI settings control.
Available to: All authenticated roles for conversations; root only for provider model, key, limits, and timeout configuration.
What this page does
AI Mode can answer operational questions, read relevant SDM state, and propose or execute permitted tool calls. It discovers the current catalog dynamically, while identity, role, zone scope, permission, and destructive confirmation remain server-side controls. It is useful for investigation and guided work; it is not a second root account with better prose.
Before you start
Ask for one clear outcome and include the zone or node when known. Review the proposed target and parameters before approving a write. Never paste passwords, API keys, 2FA codes, license tokens, provider secrets, or DNSSEC private key material into a conversation.
Models and operating limits
The current configuration UI recognizes gpt-5.6-sol, gpt-5.6-terra, gpt-5.6-luna (default), gpt-5.5, gpt-5.5-pro, gpt-5.4, gpt-5.4-mini, gpt-5.4-nano, and gpt-5.4-pro. Actual provider availability still depends on the configured account.
| Control | Current default | Meaning |
|---|---|---|
| Requests per hour / day | 60 / 240 | Bounds usage per configured policy. |
| Prompt length | 8,000 characters | Limits one submitted request. |
| Tool loops | 8 | Limits repeated inspect/act cycles in one answer. |
| Tool result | 18,000 characters | Bounds returned operational context. |
| Timeout | 90 seconds | Stops a request that does not complete in time. |
What AI Mode deliberately cannot do
- It does not receive private cluster transport, login/session, or external credential-management operations.
- Password, API-key, 2FA, license activation, direct synchronization registration, and audit-ingest credentials remain human-controlled.
- Destructive actions require explicit confirmation and non-root conversations remain role and ownership scoped.
- Conversation history belongs to the current actor and can be loaded or deleted from that actor's workspace.
Controls and fields
| Control, Field, Or Section | What It Does | What It Affects | Recommended Usage |
|---|---|---|---|
| New chat | Starts a clean assistant conversation. | Does not remove previous conversations; it creates a separate context. | Use when switching to a different customer, incident, or task. |
| Conversation search | Filters the chat list by visible conversation text. | Changes only the conversation list display. | Use to reopen previous troubleshooting work. |
| Load more | Retrieves older conversations when available. | Extends the visible history list. | Use when a previous incident is not in the initial list. |
| Prompt chips | Insert prebuilt prompts for health, DNS zones, or API access. | Helps start a structured request without typing from scratch. | Use chips for common support workflows, then add customer-specific context. |
| Message box | Accepts the operator request to the assistant. | Sends the prompt to the configured provider through SDM controls. | Be specific and avoid pasting secrets or private keys. |
| Send | Submits the current prompt. | Creates an assistant response and may include suggested API actions. | Review the response before acting on destructive recommendations. |
| AI Settings | Opens enablement, rate-limit, prompt-limit, and provider-key controls. | Controls whether AI Mode can operate and how much it can be used. | Restrict access to administrators who own provider cost and data policy. |
| Provider key controls | Save, clear, reveal, or hide the OpenAI provider key. | Controls whether AI requests can be sent to the provider. | Store only approved provider keys and clear them when AI Mode is disabled for a customer. |
| API Catalog | Shows permitted SDM API actions for the signed-in role. | Limits available execution to the current user permission scope. | Use read-only routes first during diagnosis. |
| Method/path selector | Chooses the API request the assistant or operator will run. | Determines which backend operation is executed. | Confirm method and path before sending mutating requests. |
| Query JSON / Body JSON | Structured parameters for the selected API request. | Controls exactly what the API action will receive. | Validate JSON and remove unrelated values before running. |
| Run API | Executes the selected API request from the catalog. | Can read or change SDM state depending on the selected endpoint. | Use only after reviewing parameters, permission scope, and operational impact. |
Practical guidance
- Keep AI Mode enabled only when a provider key and usage policy are approved for the customer environment.
- Treat AI output as operational guidance. Confirm destructive DNS, cluster, backup, and user actions before running them.
- Use scoped user accounts when demonstrating AI Mode to customers so the catalog mirrors their real permissions.
How to use it
- Create or open a conversation.
- Ask a specific question such as “show the DNSSEC state for example.com” or “explain why this zone is not synchronizing.”
- Read the assistant's evidence and any proposed tool operation.
- For a mutation, verify the exact target, action, and parameters before confirmation.
- Use the owning SDM page to verify the result; delete obsolete conversations when their operational context is no longer needed.
Result and next check
The assistant returns an answer grounded in catalog data available to your role, and any approved operation remains visible in normal SDM state and audit history. A refusal usually means a real permission, credential, or safety boundary—not a prompt-writing contest.
Theme color