Catchpoint Analyst

Local-first

Modern 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.

MCP
Needs key
Storage
Session
LLM
No key
Intent
performance degradation
Playbook
Performance Degradation
Cost
MCP data
Savings
80-95% token reduction

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

Credential handling
Session memory only
LLM policy
Ask first
Data scope
Tight MCP query plan
Signal sources
Catchpoint + Sonar

Controls

Choose output style and LLM behavior.

LLM Budget Policy

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.

  1. 1Parse request locally
  2. 2Select bounded diagnostic playbook
  3. 3Run narrow Catchpoint MCP calls
  4. 4Reduce raw output into evidence
  5. 5Return deterministic result

MCP Calls

The app keeps calls narrow and inspectable.

test_search-tests

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