Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.langsight.dev/llms.txt

Use this file to discover all available pages before exploring further.

​
Usage

langsight investigate [OPTIONS]

​
Options

OptionDefaultDescription
--configauto-discoveredPath to .langsight.yaml
--serverall serversFocus on a specific server
--window, -w1hLook-back window (30m, 2h, 1d)
--jsonfalseOutput raw evidence JSON

​
What it does

Gathers evidence from stored health history — DOWN counts, DEGRADED events, schema drift, error messages — and sends it to an LLM for root cause analysis. Evidence gathered:
  • Health check history (last N results in window)
  • Schema drift events
  • Consecutive failure counts
  • Error message patterns
  • Average latency vs. baseline

​
AI providers

Configure your preferred provider in .langsight.yaml: 
investigate:
  provider: gemini           # anthropic | openai | gemini | ollama
  model: gemini-2.0-flash
Each provider requires the corresponding API key environment variable:
ProviderRequired env var
anthropicANTHROPIC_API_KEY
openaiOPENAI_API_KEY
geminiGEMINI_API_KEY
ollamaNone (local) — set base_url in config
# Example: use Anthropic Claude for AI-powered analysis
export ANTHROPIC_API_KEY=sk-ant-...
langsight investigate --server jira-mcp --window 2h
Fallback: When no provider is configured, or when the required API key environment variable is not set, langsight investigate falls back to rule-based heuristic analysis automatically. The rule-based engine works offline with no API key and covers the most common failure patterns (connection failures, auth errors, schema drift). The output is less detailed than AI-powered analysis but always available. See AI Providers for per-provider setup and model selection.

​
Example (with AI)

langsight investigate --server jira-mcp --window 2h

Analysing with Gemini  gemini-2.0-flash...

╭─ Root Cause Analysis  (Gemini gemini-2.0-flash) ──────────────────╮
│                                                                    │
│ ## jira-mcp                                                        │
│                                                                    │
│ **Root Cause**: Authentication token expired or revoked.           │
│                                                                    │
│ **Evidence**: 8/8 checks failed with "timeout after 5s" starting   │
│ at 14:32 UTC. No schema drift. Latency history shows normal        │
│ response times until the failure began — consistent with auth      │
│ failure rather than network or process issues.                     │
│                                                                    │
│ **Impact**: All agents using jira-mcp for ticket lookup and        │
│ updates are failing silently.                                      │
│                                                                    │
│ **Recommended Actions**:                                           │
│ 1. Check the Jira API token in the server's .env file              │
│ 2. Verify the token hasn't expired in the Atlassian console        │
│ 3. Rotate the token and restart the MCP server                    │
╰────────────────────────────────────────────────────────────────────╯

​
Example (rule-based fallback)

# No GEMINI_API_KEY set
langsight investigate

╭─ Root Cause Analysis  (rule-based) ───────────────────────────────╮
│ ## jira-mcp                                                        │
│                                                                    │
│ **Root Cause**: Server unreachable — connection failure.           │
│ **Evidence**: 8/8 checks failed (100%). Error: timeout after 5s   │
│ **Actions**:                                                       │
│ 1. Check if the MCP server process is running                      │
│ 2. Verify the command path in .langsight.yaml                      │
│ 3. Run langsight mcp-health for current status                     │
╰────────────────────────────────────────────────────────────────────╯