From Signals to Prediction
Lightweight Agent for AI-Powered PostgreSQL Intelligence
Distribution Repository — This repository contains release binaries, installation scripts, and documentation.
Source code is closed-source. For licensing inquiries: licensing@burnsideproject.ai
Install | Features | AI Prediction | Tiers | Quick Start | Configuration
What is pg-collector?
pg-collector is a lightweight edge compute agent that extracts PostgreSQL telemetry and delivers it to configurable destinations—locally for evaluation, or streamed to cloud platform where AI LLM analyzes patterns and predicts issues before they impact your users.
The 5 answers as first-class product outputs
- "Is my database healthy?”
- “What changed?”
- “What will break next?”
- “What caused this spike?”
- “How is performance trending over 30 days?”
Why pg-collector is not just another observation tool?
pg-collector backend continuously distill raw database telemetry into long-running trend analysis features. The longer a customer is on the platform, the richer those trends become — and the smarter our AI reasoning agent gets. The data grows, but the intelligence compounds.
What is the secret sauce
Deep, high-fidelity PostgreSQL telemetry collection — built lean, engineered for intelligence, and optimized for AI-driven diagnostics.
Design Principals
- Act like an IoT sensor - Lightweight edge device that observes and transmits, never interferes.
- Never lose data - Samples flow through an in-memory ring buffer into DuckDB. If DuckDB is busy, samples spill to SQLite and drain back automatically.
- Resilient by default — Works offline with local storage for up to 72 hours.
- Zero Trust - No UID/Password. Only mTLS, AWS/GCP IAM required
- Single binary, zero dependencies — No JVM, no Python runtime, no agents to install. One YAML config, one binary, runs anywhere — systemd, Docker, Kubernetes, bare metal.
- Intelligence compounds over time — Raw telemetry feeds long-running trend analysis. The longer you collect, the richer the patterns, the smarter the AI predictions become.
┌──────────────┐ ┌──────────────┐ ┌──────────────┐ ┌──────────────┐
│ PostgreSQL │ ───▶ │ pg-collector │ ───▶ │ Multi or | ───▶ | Predictions |
| Database | | Edge (Agent) | | Single Tenant| | & Alerts |
| │ │ │ | AWS or GCP | | |
│ │ │ │ │ Cloud │ │ |
└──────────────┘ └──────────────┘ └──────────────┘ └──────────────┘
pg-collector is available in two editions:
Try it locally—no cloud account required.
- Single binary - zero external dependencies
- Runs locally - all output stays on your machine
- Cut and Paste - the outputs from pg-collector to your local AI Chat Windows
- Core PostgresSQL - samplers activity, database, and statements (PostgresSQL Wiki)
- Example Prompt - is provided for instant analysis
- Perfect for evaluation - learning, and local deployment
Everything in Demo, plus the full power of AWS or GCP Cloud services
- Up to 12 PostgreSQL metric samplers — activity, performance, replication, WAL, locks, vacuum, bloat, and more
- AI-powered health reports - with prescriptions and root cause analysis
- No Root Account - just "pg_monitor"
- No special function or sp - we use native postgres observability tools (PostgreSQL Wiki)
- Intentionally designed - non-intrusive in your current postgres
- Predictive analytics - anomaly detection
- Real-time streaming - to AWS + GCP Multi-Tenant or Single-Tenant Cloud
- Interactive health dashboard - with configuration audit
- PII detection - audit logging, and query masking
- Multi-database monitoring — scale from one database to unlimited
- Offline resilience — continues collecting during network outages
- Enterprise authentication — mTLS, AWS IAM, GCP IAM Only
- Flexible deployment — systemd, Docker, Kubernetes
- Dedicated support — email, priority, and SLA tiers available
Ready to upgrade? Book a demo or visit burnsideproject.ai.
Want to try AI-powered analysis without the cloud?
Download Standalone Local Demo
This standalone local testing tool that lets you analyze pg-collector telemetry using your own LLM (ChatGPT, Claude, etc.). No API keys or cloud infrastructure required.
┌─────────────────────────────────────────────────────────────────────────┐
│ Your Environment (Edge Only) │
│ │
│ PostgreSQL ──→ pg-collector ──→ ./telemetry/*.jsonl │
│ (single binary) │ │
│ ▼ │
│ ./prepare-snapshot.sh │
│ │ │
│ ▼ │
│ telemetry-snapshot.json ──→ Your LLM │
└─────────────────────────────────────────────────────────────────────────┘
Follow the README.md enclosed in the download
Test with your own LLM Chat Application (ChatGPT , Claude - We provide you demo Prompt but feel free to use your own prompt)
| Prompt | Use Case |
|---|---|
01-health-check.md |
General database health assessment |
02-slow-queries.md |
Query performance optimization |
03-incident-investigation.md |
Diagnose active problems |
04-capacity-planning.md |
Resource utilization and scaling |
05-quick-diagnosis.md |
Fast 30-second triage |
| Capability | Demo (Free) | Commercial (Subscription) |
|---|---|---|
| Single binary, runs locally | ✅ | ✅ |
| Core samplers (activity, database, statements) | ✅ | ✅ |
| Export snapshots to your LLM | ✅ | ✅ |
| Real-time cloud streaming | ❌ | ✅ |
| Continuous monitoring | ❌ | ✅ |
| AI health reports with prescriptions | ❌ | ✅ |
| Interactive health dashboard | ❌ | ✅ |
| Configuration audit | ❌ | ✅ |
| Predictive analytics & anomaly detection | ❌ | ✅ |
| Extended samplers (WAL, locks, replication, bloat, etc.) | ❌ | ✅ |
| Automated alerts (Slack/PagerDuty) | ❌ | ✅ |
| Historical analysis | ❌ | ✅ |
| Multi-database monitoring | ❌ | ✅ |
Download the latest release from the Releases page.
| Platform | Architecture | File |
|---|---|---|
| Linux | x86_64 (amd64) | pg-collector-linux-amd64.tar.gz |
| Linux | ARM64 | pg-collector-linux-arm64.tar.gz |
| macOS | Intel (x86_64) | pg-collector-darwin-amd64.tar.gz |
| macOS | Apple Silicon | pg-collector-darwin-arm64.tar.gz |
| Windows | x86_64 | pg-collector-windows-amd64.zip |
Each release includes a checksums.txt file. After downloading:
sha256sum -c checksums.txt --ignore-missingStart small, scale as you grow. All commercial plans include predictive analytics and alerting.
| Starter | Pro | Business | Enterprise | |
|---|---|---|---|---|
| Databases | 1 | Multiple | Many | Unlimited |
| Metric Samplers | Core | Extended | Full | Full + Custom |
| Alert Frequency | Daily digest | Near real-time | Real-time | Real-time |
| Data Retention | Standard | Extended | Long-term | Custom |
| Query Masking | Basic | Full | Full | Custom rules |
| AI Insights | Summary reports | Detailed analysis | Predictive | Interactive + Custom |
| Support | Community | Priority | SLA-backed |
Ready to get started? Book a demo or try the free evaluation.
The intelligence behind the insights. Our AI transforms raw database metrics into predictions you can act on.
┌─────────────┐ ┌─────────────┐ ┌─────────────┐
│ PG Collector│ ───▶ │ AWS or GCP │ ───▶ │ Alerts │
│ Metrics │ | AI Agents | | |
| | | Predictions│ │ Slack/Email │
└─────────────┘ └─────────────┘ └─────────────┘
| Prediction Type | What It Detects |
|---|---|
| Connection Exhaustion | Pool approaching limits based on growth patterns |
| Replication Lag Spike | Replica falling behind due to write surge |
| Lock Contention | Blocking chains forming from concurrent transactions |
| Cache Pressure | Buffer cache hit ratio degrading |
| Vacuum Emergency | Tables approaching transaction wraparound |
| Query Degradation | Execution plans changing, slow query emergence |
The AI engine continuously analyzes your PostgreSQL metrics, detects patterns, and delivers:
- Severity-rated alerts — Know what's critical vs. informational
- Confidence scoring — Understand how certain the prediction is
- Evidence-based reasoning — See the data points driving the alert
- Recommended actions — Get specific, actionable fix suggestions
When connected to Burnside Project Cloud Engine, commercial subscribers get access to:
Control Panel
| Feature | Description |
|---|---|
| Health Dashboard | Real-time overview of all monitored databases with status indicators |
| AI Health Reports | Automated database health assessments with severity ratings |
| Prescriptions | Specific, actionable fix recommendations generated by AI |
| Configuration Audit | Detects suboptimal PostgreSQL settings and suggests improvements |
| Anomaly Detection | Machine learning identifies unusual patterns in your metrics |
| Rule Engine | Customizable alert rules with flexible thresholds and conditions |
| Query Analytics | Statement-level performance tracking with trend analysis |
| Replication Monitoring | Lag tracking across primary and replica topologies |
| Vacuum Intelligence | Proactive alerts for autovacuum health and wraparound risk |
| Capacity Planning | Growth trend projections for connections, storage, and throughput |
| Historical Analysis | Query and compare metrics across configurable retention windows |
| Multi-Database Views | Unified monitoring across all your PostgreSQL instances |
Connects to Control Plane like an Edge Device
mTLS authentication to your control plane
Analyze what query is creating contention
AI-Generated Prescriptive Runbooks for Faster Resolution
Predictive Risk Score from predictive ML model
Predictive Risk Score derived from grounding truth
Daily Predictive Risk Score derived from grounding truth
Building State Machine from grounding truth
Al Analysis
What is loaded vs Recommended
Complete matrix rolled up hourly
Aggregated Tiles (Hourly)
Daily Prediction with grounding truth
Detail Metrics
Current Config Audit + Recommendation with Postgres Recommended Alignments
Cloud Mico Batching Pipeline
Building Adaptive State
Feature Store that AI can label Normal vs Outside Normal
Features building defining 950, p95, p99
| Category | What You Learn | How Fast |
|---|---|---|
| Activity | Which queries are running, waiting, or blocked | Real-time |
| Performance | Slow queries, cache misses, I/O bottlenecks | Continuous |
| Replication | Lag alerts before replicas fall behind | Real-time |
| Storage | Table bloat and growth trends | Periodic |
| Background | Vacuum health and checkpoint pressure | Periodic |
| Guarantee | What It Means |
|---|---|
| Memory Safe | Configurable ceiling with automatic management—never runaway |
| Network Resilient | Hybrid ingestion pipeline with SQLite overflow survives outages—zero data loss |
| Minimal Footprint | Max 2 PostgreSQL connections, timeout-protected queries |
| Secure by Default | mTLS or IAM auth—no passwords in config files |
| Method | Best For | Guide |
|---|---|---|
| mTLS Certificates | Self-managed PostgreSQL (most secure) | Security Guide |
| AWS IAM | RDS, Aurora (passwordless) | AWS Setup |
| GCP IAM | Cloud SQL (passwordless) | GCP Setup |
You configure your API key and database connection. Metrics delivery is handled automatically.
# Your API key (provided during onboarding)
api_key: "${API_KEY}"
# Your PostgreSQL connection
postgres:
conn_string: "postgres://pgcollector@your-db:5432/postgres?sslmode=verify-full"
auth_method: cert
tls:
mode: verify-full
ca_file: /etc/pg-collector/certs/ca.crt
cert_file: /etc/pg-collector/certs/client.crt
key_file: /etc/pg-collector/certs/client.keySee Configuration Guide for all options.
sudo systemctl enable pg-collector
sudo systemctl start pg-collector
sudo systemctl status pg-collectordocker run -d \
--name pg-collector \
-v /etc/pg-collector:/etc/pg-collector:ro \
ghcr.io/burnside-project/pg-collector:latest# Basic health check
curl http://localhost:8080/health
# Detailed status
curl http://localhost:8080/status
# Prometheus metrics
curl http://localhost:8080/metricsFrom install to insights in under 5 minutes.
- Install — Download and run the single binary
- Connect — Point it at your PostgreSQL database
- Stream — Metrics flow securely to Burnside Cloud
- Predict — AI analyzes patterns and alerts you before problems hit
| Guide | Description |
|---|---|
| Quick Start | Get running in 5 minutes |
| Installation | Detailed installation guide |
| Configuration | All configuration options |
| Security | mTLS setup, certificate management |
| AWS Setup | RDS, Aurora deployment |
| GCP Setup | Cloud SQL deployment |
| Monitoring | Health checks, Prometheus metrics |
| CLI Reference | Command-line options |
| Troubleshooting | Common issues and solutions |
| FAQ | Frequently asked questions |
For detailed commercial documentation, see the /docs directory.
- Documentation: docs/
- Issues: GitHub Issues
- Sales: sales@burnsideproject.ai
- Support: support@burnsideproject.ai
Copyright © 2024–2026 Burnside Project LLC. All rights reserved.
This software is licensed under the Burnside Project Community Freeware License v1.1. This is not open source software.
You may use this software free of charge for personal, educational, evaluation, and private business purposes.
You may not redistribute, sell, sublicense, or offer this software as a hosted service, SaaS, or component of a commercial product without a separate commercial license.
For commercial licensing inquiries, contact licensing@burnsideproject.ai.
See LICENSE for the complete license text
From Signals to Prediction. Stop reacting. Start predicting.
AI-powered database observability by Burnside Project