Catchpoint Analyst
Local-firstModern monitoring intelligence with cost controls built in.
Use deterministic playbooks for repeatable analysis, add optional client context for reports, and reserve LLM calls for compact evidence packs.
Answer
Dynamic result from the selected playbook.
Plan the question or run MCP to generate a dynamic answer from live Catchpoint data.
Evidence Pack
84% playbook confidence
Controls
Choose output style and LLM behavior.
Decision Guardrail
When to spend on an LLM.
Use local rules for repeatable analysis and evidence shaping.
Use LLM only for customer-ready wording, prospect personalization, or ambiguous cause ranking.
Never send raw payloads; send the answer, evidence pack, and context fields.
Execution Plan
Deterministic path before any optional model call.
- 1Parse request locally
- 2Select bounded diagnostic playbook
- 3Run narrow Catchpoint MCP calls
- 4Reduce raw output into evidence
- 5Return deterministic result
MCP Calls
The app keeps calls narrow and inspectable.
Resolve synthetic tests before performance comparison.
{
"name": "checkout"
}Report Draft
Generated locally; LLM polish is optional.
Triage Summary Client: Not specified Industry: Not specified Prepared for: General technical audience Audience: Internal / customer-safe Window: Last 24 hours Service focus: Derived from question Objective: Identify impact, likely cause, and next action. Sources: Catchpoint + Sonar Question Why is checkout slow today? Executive Summary The request was classified as performance degradation and routed to the Performance Degradation playbook. The current evidence pack is compact enough to use for a deterministic summary or an economical LLM polish pass. Evidence - Credential handling: Session memory only - LLM policy: Ask first - Data scope: Tight MCP query plan - Signal sources: Catchpoint + Sonar Findings - This request maps to a repeatable diagnostic path. - Filtering, grouping, ranking, thresholds, and report scaffolding stay local. - LLM use is blocked for this run. Recommended Actions - Review the planned MCP call before running it. - Normalize returned data into local scorecards. - Use LLM only for evidence-pack narration when it saves human writing time. LLM Budget Policy Ask before any LLM call. If approved, send only this compact evidence pack.
Findings
Local interpretation rules.
- This request maps to a repeatable diagnostic path.
- Filtering, grouping, ranking, thresholds, and report scaffolding stay local.
- LLM use is blocked for this run.
Playbook Library
Bounded patterns, not infinite prompting.
Availability Drop
failures, affected regions, error mix, Sonar correlation
Performance Degradation
response time, wait, DNS, SSL, load, baseline delta
RUM Impact
page views, sessions, bounce, conversion, geography, device
Network / Sonar
outage score, ASN, regions, packet loss, BGP events