Open Source · PostgreSQL · MIT License

See which tenant is
killing your database

Tenant-aware observability and autonomous control for multi-tenant PostgreSQL. Detect noisy neighbors, attribute costs, and auto-throttle — in a single binary.

Get Started — Free Quick Start →
FaultWall Dashboard — tenant leaderboard, cost attribution, anomaly detection
# One binary. One env var. Done.
DATABASE_URL=postgres://... ./faultwall
# → Dashboard at localhost:8080
5,000+
Lines of Go
10
MCP Tools
1
Dependency
<1%
Overhead
Features

Everything your database can't tell you

PostgreSQL sees queries. FaultWall sees tenants.

🔍

Auto-Detection

Automatically identifies your tenant isolation pattern — schema-per-tenant, row-level, or database-per-tenant. Zero config needed.

🧠

Anomaly Detection

Statistical learning builds per-tenant baselines. Z-score analysis flags deviations. No LLM, no API keys — runs locally.

🔮

Predictive Alerts

Linear regression on metric trends: "acme_corp will breach threshold in ~4 minutes." Act before the outage.

Auto-Throttling

Kill runaway queries, enforce per-tenant connection limits. Configurable grace periods and cancel → terminate escalation.

💰

Cost Attribution

"Tenant X costs $338/mo of your $360 RDS bill." Proportional allocation based on actual query time consumption.

🤖

AI-Native (MCP)

10-tool MCP server lets AI agents detect noisy neighbors and throttle them autonomously. No human in the loop.

How it works

Four steps to tenant visibility

1

Point it at your database

Set DATABASE_URL and run the binary. FaultWall connects as a read-only client — it never modifies your data.

2

It auto-detects your tenants

Polls pg_stat_statements every 10 seconds, identifies your isolation pattern, and starts building per-tenant metrics.

3

See who's causing problems

Real-time dashboard: tenant leaderboard, cost attribution, anomaly alerts, and breach predictions. Auto-refreshing.

4

Auto-throttle the noisy neighbors

FaultWall auto-kills runaway queries and enforces limits. Or let your AI agent handle it via MCP.

AI-Native

Built for autonomous agents

Every monitoring tool has a dashboard for humans. FaultWall is the first natively controllable by AI.

You: Which tenants are causing problems?
→ Claude calls get_noisy_tenants
Claude: acme_corp has avg query time 234ms and is using 74% of resources ($338/mo). Their analytics export started 3 minutes ago. Want me to throttle them?
You: Yes, do it.
→ Claude calls throttle_tenant("acme_corp", "cancel")
Claude: Done. Cancelled 3 active queries. P99 latency already dropping for other tenants.
Deep Tech

Kernel-level resource attribution

What you can't see, you can't control. We went deeper than anyone else.

🛡️

Read-Only. Zero Impact on Your Queries.

FaultWall connects as a read-only client. It polls Postgres catalog views (pg_stat_statements, pg_stat_activity) every 10 seconds — the same views every monitoring tool reads. It never modifies your data, never adds queries to your workload, and never touches your application traffic.

Read-only connection
No query interception
No schema changes
<1% overhead (eBPF)

eBPF Engine

We hook directly into the Linux kernel's scheduler and block I/O subsystem using eBPF. Every CPU nanosecond and every disk byte is attributed to the exact PostgreSQL PID — and mapped back to the tenant in real-time.

Why this matters

Traditional tools poll aggregated stats. FaultWall traces at the kernel level: "This INSERT from tenant acme_corp used 12ms of CPU and 480KB of disk I/O." No other PostgreSQL monitoring tool provides this precision.

🧬 AutoResearch — Self-Tuning Detection

FaultWall deploys an autonomous optimization loop inspired by Andrej Karpathy's AutoResearch. A genetic algorithm continuously evolves detection parameters — sensitivity thresholds, window sizes, baseline intervals — against your real workload.

60%
Before tuning
100 generations
100%
After tuning · 0 false positives

The longer FaultWall runs, the better it understands your database. No LLM needed — pure statistical optimization in Go.

Want kernel-level tenant attribution?

Our eBPF engine is available for teams running self-hosted PostgreSQL. Validated on Linux 5.8+ with PostgreSQL 14-16.

Contact Us →

shreyas@faultwall.com

Your database is on fire.
FaultWall tells you who lit the match.

Open source. MIT licensed. One binary. Zero config.

Get Started — it's free →