How It Works

Type a question in plain English. The Denali reads your question, decides whether to use a named analytics capability or write custom SQL, executes the query against Redshift, and returns a natural-language summary alongside the raw result table.

Responses are async — the chat polls for the result every 2 seconds. Most queries complete in 5–30 seconds depending on data volume.

Model

• Provider: Amazon Bedrock — Anthropic Claude
• Model: Claude Sonnet 4.6 (us.anthropic.claude-sonnet-4-6) via US cross-region inference profile
• Context window: 200,000 tokens — the full schema, all 20 capability templates, and your conversation history fit comfortably
• Max result rows: 5,000 per query (enforced server-side)
• Conversation memory: The last 10 turns are sent with every message so you can ask follow-up questions

System Prompt (what the Denali knows)

Every request includes a system prompt with:

• Full schema for diality_spectrum.processed_logs, payload_details, alarm_details, and all 14 legacy views in diality_logs
• All 20 named capability templates (Treatment Summary, Alarm Events, DG Pressure Graph, Cross-Device Comparison, etc.)
• Field definitions: DeviceId, hex_id, source (HD/DG), alarm_name, priority, date (yyyyMMdd), Datetime (timestamp)
• Safety rules: SELECT-only queries, allowed schemas only, LIMIT enforced

The Denali always responds with a JSON action (run_capability, run_sql, or answer) — never raw prose — so every data response is backed by an auditable SQL query you can expand with View SQL.

Safety & Limits

• Only SELECT / WITH queries are allowed — no INSERT, UPDATE, DELETE, DROP, or DDL
• Only diality_spectrum and diality_logs schemas are accessible
• All queries are capped at 5,000 rows — a truncation note appears when hit
• Results are never stored beyond 24 hours (DynamoDB TTL)

Keyboard Shortcuts

• Enter — Send message
• Shift + Enter — New line in your message

Sample Prompts

Simple

• How many files have been processed?
• What devices have uploaded logs?
• Show me the most recent alarm events.
• What are the HIGH priority alarm names?
• How many log lines were processed per day this week?

Intermediate

• Show me all HIGH priority alarms for device DVTHD021 in the last 30 days.
• Compare processed row counts across all devices for last month.
• Show the DG pressure trend for DVTHD021 on 20251029.
• Which alarm IDs triggered most frequently across all devices?
• Show me the treatment summary for all devices last week.
• What files failed processing and what were the errors?

Complex

• Which devices had more than 10 HIGH priority alarms in a single day, and on which dates?
• Show me the top 5 most frequent alarm names per device and compare them side by side.
• For each device, how many distinct hex_id message types were seen, and which device had the most variety?
• Show me heat disinfect cycles that were interrupted by alarms — include the device, date, and alarm name.
• Which days had both a high alarm count and a long average processing duration? Show device, date, alarm count, and duration.

Tips

• You can ask follow-up questions — the Denali remembers the last 10 turns
• Mention device IDs (DVTHD021), dates (20251029), alarm names, or hex IDs to narrow results
• Click View SQL under any response to see the exact query that ran
• Click Clear conversation to start fresh with a new topic
• If the Denali misunderstands, rephrase with more specific field names (e.g. "alarm_name", "DeviceId", "date")