AI Model: Llama 3.1

Use case: Natural language time range generator to turn phrases like “Tuesday last week,” “Black friday,” or “last night” into a real Observe by Snowflake time range.

Where it's available: The global time-range picker in explorers, worksheets, and dashboards.

LLM invocation: Only if the local parser cannot resolve the string. In such cases, Llama 3.1 70B (Snowflake Cortex) maps the string to a preset or exclusive ISO range in the user’s timezone. Structured inputs like “Past 15 minutes” don't hit the model.

AI Model: GPT 5.1

Use case: Product and documentation Q&A, such as “how do I write this OPAL?”, or “where is X?”. AI Help does not investigate your telemetry.

Where it's available: Help > AI Help from the left navigation. Disabled if you turn off external AI from your Workspace settings.

LLM invocation: On each question, the bot accesses Observe by Snowflake documentation, generates answers (GPT-5.1 via Cortex in the current routing), and a second pass scores/filters weak or broken-link replies. It does not query customer Datasets.

AI Model: GPT 5.1

Use case: Turn messy log/string samples into an extract_regex with named capture groups (new columns).

Where it's available: From a string cell or column in Log Explorer, Worksheets, or a Dataset pipeline table, choose Extract from string. Configure capture groups and the optional prompt under Use o11y AI to help. Requires external AI to be enabled.

LLM invocation: Opening the modal asks the model for suggested capture groups. Clicking Generate regex sends samples, selected capture groups, and the optional prompt to Cortex GPT-5.1 and returns a regex plus a short explanation. Editing the regex box yourself does not call the model.

AI Model: Sonnet family

Use case: Ghost-text completions while writing OPAL, such as next tokens or lines. This is not a chat that designs a whole pipeline.

Where it's available: Open the OPAL tab and begin typing. When the grey auto-completions appear, press Tab to accept the suggestion.

LLM invocation: Press Command (⌘) + K in macOS or Cmd + K in Windows in the OPAL editor window.

AI Model: Opus family

Use case: Natural language chat to investigate incidents, search Datasets, write queries, and auto-triage when a Monitor fires.

Where it's available: From the left navigation rail, the Home page, from an alert, various side panels across product areas.

LLM invocation: On every user message and alert-triggered investigation. AI SRE loops with tools such as Dataset search, query cards, and skills until an answer is obtained. This is the heavy Opus path with high reasoning, not a single completion.

AI Model: Opus family

Use case: Describe the alert you want, then the AI assistant configures a v2 Monitor with the correct configuration such as type, query, thresholds, noise, and missing fields.

Where it's available: When creating or editing Monitors.

LLM invocation: When you send a prompt or land with an initial prompt from the create screen. The AI agent reads the current Monitor definition and uses page tools to propose or apply changes. This is the Opus/high-reasoning path so it can reason about query cost and alert quality, not a one-shot form fill.

AI Model: Sonnet family

Use case: Explain a single distributed trace: overview, service call graph, bottlenecks (N+1, slow DB/cache), errors, and instrumentation gaps.

Where it's available: Open a trace, then click Summarize trace.

LLM invocation: When you open the panel or send a prompt. The page sends the current waterfall (capped span set) as context; Sonnet answers against that snapshot. It does not roam the rest of the workspace the way full AI SRE does.