TriageProf - AI-Powered Performance Profiling

TriageProf is a modular profiling and bottleneck-analysis tool for Go applications. It collects pprof profiles from a live service, runs deterministic analysis, then enriches findings with Mistral AI to explain why things are slow and exactly what to fix — producing a beautiful, self-contained HTML report.

🚀 Quick Start

Installation

# Clone the repository
git clone https://github.com/lsanarchist/Mistral-Hackathon.git
cd Mistral-Hackathon

# Build the binary and plugins (single step)
make build

# List available plugins
./bin/triageprof plugins

Run Against Your Own Service

# Profile a live Go service and get an AI-powered report
export MISTRAL_API_KEY="your-key-here"

./bin/triageprof run \
  --plugin go-pprof-http \
  --target-url http://localhost:6060 \
  --duration 15 \
  --outdir ./analysis \
  --llm

# After the run, triageprof will offer to open the report in your browser automatically

✨ Features

Core Capabilities

• 🔍 Comprehensive Profiling: CPU, heap, allocation, block, and mutex profiling
• 🤖 Deterministic Analysis: 8+ rule-based patterns, no LLM required for findings
• 🧠 Mistral AI Enrichment: mistral-large-latest explains root causes, suggests fixes, estimates effort
• 📊 Self-contained HTML Report: Interactive dark-theme report — charts, hotspot bars, per-finding AI cards
• 🌐 Auto Browser Serve: After each run, offers to serve the report and open it in your browser
• 🔌 Plugin Architecture: Extensible via JSON-RPC plugin SDK

Analysis Rules

1. CPU Hotpath Dominance: Detects functions consuming CPU time
2. Allocation Churn: Identifies high mallocgc/memmove patterns
3. JSON Hotspots: Finds encoding/json bottlenecks
4. String Churn: Analyzes strings.Builder/bytes.Buffer usage
5. GC Pressure: Measures runtime.gcBgMarkWorker impact
6. Mutex Contention: Detects sync.(*Mutex).Lock issues
7. Heap Allocation: Identifies memory allocation hotspots
8. Block Contention: Analyzes runtime.chan/select patterns

Report Contents

• Score gauge — overall health at a glance
• Severity breakdown — donut chart + filter by critical / high / medium / low
• AI Executive Summary — Mistral's verdict with confidence score and key themes
• Top Risks — what will blow up under load
• Recommendations — prioritized actions with effort, complexity, code examples, validation metrics
• Per-finding AI cards — root causes ↔ suggestions, next measurements ↔ caveats, before/after metrics

🔧 Configuration

LLM Options

export MISTRAL_API_KEY="your-api-key"

./bin/triageprof run \
  --plugin go-pprof-http \
  --target-url http://localhost:6060 \
  --outdir ./analysis \
  --llm \
  --llm-model mistral-large-latest \
  --llm-timeout 90

Flag	Default	Description
--llm	off	Enable Mistral AI enrichment
--llm-model	mistral-large-latest	Model to use
--llm-provider	mistral	mistral or openai
--llm-timeout	20	API timeout in seconds

🎯 Commands

# Full pipeline: collect → analyze → AI enrich → HTML report
./bin/triageprof run --plugin go-pprof-http --target-url <url> --outdir <dir> [--llm]

# Just collect profiles into a bundle
./bin/triageprof collect --plugin go-pprof-http --target-url <url> --out bundle.json

# Analyze an existing bundle
./bin/triageprof analyze --in bundle.json --out findings.json

# Generate LLM insights from findings
./bin/triageprof llm --bundle bundle.json --findings findings.json --out insights.json

# List plugins
./bin/triageprof plugins

📦 Output Structure

analysis/
├── report.html            # Self-contained interactive HTML report (~300KB)
├── index.html             # Alias for report.html (same content)
├── findings.json          # Structured performance findings
├── insights.json          # Mistral AI analysis (if --llm enabled)
├── report.md              # Markdown report
└── bundle.json            # Raw collected profiles bundle

🔌 Plugin System

Profilers are separate executables communicating via JSON-RPC over stdio. The go-pprof-http plugin is included and works out of the box for any Go service with net/http/pprof enabled.

# List available plugins
./bin/triageprof plugins list

# Your service just needs this import:
import _ "net/http/pprof"

See API Documentation for the plugin development guide.

📚 Documentation

• User Guide: Complete usage guide with examples
• CLI Reference: Detailed command reference
• API Documentation: Plugin SDK and JSON-RPC API
• Contributing Guide: Development and contribution guidelines

🤝 Contributing

• Issues: Report bugs and request features
• Pull Requests: Contributions welcome!

See CONTRIBUTING.md for guidelines.

📜 License

TriageProf is licensed under the MIT License.

---

TriageProf — built for the Mistral AI Hackathon 🚀