omx agent

Ask OnlyMetrix a data question. The agent runs deterministic CLI commands to investigate, then narrates the findings.

The LLM picks which tools to run. The Rust engine executes them. Every step is auditable.

Usage

omx agent "why did revenue drop 15%?"
omx agent "what metrics are failing?" --explain
omx agent "check reliability" --json

Flags

Flag	Description
--explain	Show each investigation step with tool name, summary, and duration
--json	Machine-readable JSON output

Example: Investigation Mode

omx agent "why did revenue drop?" --explain

Running investigation...

  Step 1: reliability_check (50ms)
    > 6/11 healthy, 5 failing, 5 degraded

  Step 2: query_metric (9ms)
    > quarterly_revenue — 9 rows

  Step 3: query_metric (7ms)
    > revenue_by_country — 10 rows

  Step 4: query_metric (3ms)
    > total_revenue = 17.7M

  Step 5: run_analysis (12ms)
    > analysis complete

Answer:
Revenue metrics are degraded due to 7,901 invoices with negative
total amounts. Key metrics affected: quarterly_revenue,
revenue_by_country, total_amount, and total_revenue.

Inspect: SELECT * FROM invoices WHERE total_amount < 0 LIMIT 10

  Commands run: 5
  Session: 120b8bca-f9c1-4b3d-9244-08b1b491b92e

How It Works

1. Your question goes to POST /v1/chat with investigation_mode: true
2. The Rust agent runs a tool loop (max 4 turns)
3. Available tools: query_metric, list_metrics, reliability_check, run_analysis
4. Each tool call is executed in-process by the Rust engine — no SQL generation
5. The LLM composes a narrative from the structured results
6. Every step is returned in the steps[] array with tool name, args, summary, and duration

JSON Output

omx agent "check reliability" --json

{
  "type": "finding",
  "narrative": "6 of 11 metrics are healthy...",
  "steps": [
    {
      "tool": "reliability_check",
      "args": {},
      "summary": "6/11 healthy, 5 failing",
      "duration_ms": 48
    }
  ],
  "session_id": "abc-123"
}