GEARFLOW LAB // CENTRAL ENGINEERING TOOLS DIRECTORY 10 Client-Side Utilities · 34 Live Presets · 10 PDF Runbooks · Zero Ads

GearFlow Lab Engineering Calculators & Config Generators

Validate database memory ceilings, reverse-proxy SSE timeouts, hardened Docker Compose files, host RAM and ZRAM budgets, quantized LLM VRAM, and macOS 5K2K display bandwidth locally in your browser. Every utility runs 100% client-side with zero server uploads and deterministic shareable URLs.

Showing 10 of 10 utilities
FLAGSHIP DIAGNOSTIC GATEWAY // 60 PRODUCTION SIGNATURES · < 5 MS LOCAL PARSE

Instant DevOps & AI Stack Error Log Decoder

NEW FLAGSHIP · 60 SIGNATURES · CASCADE TRACE 100% Browser-Local · Zero Server Upload

Docker, Nginx, PostgreSQL, n8n & Ollama Error Log Decoder (100% Client-Side)

Paste multi-line docker compose logs, nginx error.log, PostgreSQL / PgBouncer, n8n / BullMQ / Redis, or Ollama / vLLM traces in your browser. Isolates cascading multi-error root causes (such as Exit Code 137 triggering downstream HTTP 502), outputs copy-ready read-only diagnostic CLI + zero-downtime fixes, and opens the exact prefilled sizing calculator below.

CATEGORY 01 // CONNECTION POOLS & STREAMING PROXIES

Database & Reverse Proxy

NEW · NGINX · CADDY 2 · TRAEFIK V3 · CLOUDFLARED 4 Engines · 127.0.0.1 UFW Safe

Nginx, Caddy 2, Traefik v3 & Cloudflare Tunnel Config Matrix Generator

Generate side-by-side production reverse-proxy and QUIC tunnel configs across 6 workloads (n8n MCP SSE, Ollama Bearer auth, OpenObserve, FastAPI WebSockets, PgBouncer TLS TCP, and WordPress PHP-FPM) with automatic 127.0.0.1 UFW loopback bindings and live curl -i -N checks.

POSTGRESQL 16/17 · PGBOUNCER postgresql.conf + pgbouncer.ini

PostgreSQL & PgBouncer Memory/Pool Calculator

Size RAM-safe shared_buffers, worst-case work_mem, max_connections, and PgBouncer transaction mode pools (max_prepared_statements = 100) to eliminate SQLSTATE 53300 and cgroup OOM crashes.

NGINX · MCP SSE · PHP-FPM nginx.conf + http://www.conf

Nginx 502/504 & MCP SSE Timeout Config Generator

Generate coordinated proxy_read_timeout hierarchies, upstream header buffers, unbuffered proxy_buffering off blocks for n8n MCP and Ollama token streams, and RAM-bounded PHP-FPM worker pools.

CATEGORY 02 // CONTAINER HARDENING, CGROUP RAM & TCO

Docker, Compose & VPS Sizing

NEW · SYSTEMD · LAUNCHD · UNHEALTHY AUTO-HEAL 4-Tab Deadlock Watchdog

Systemd, macOS LaunchAgent & Docker Healthcheck Deadlock Watchdog Generator

Standalone restart: unless-stopped never restarts a wedged PID 1 container stuck in unhealthy. Generate coordinated IPv4 127.0.0.1 Compose healthchecks, Linux systemd + docker events auto-healers, macOS LaunchAgent plists, and SIGSTOP fault-injection drills.

NEW · STACK RAM · ZRAM · VPS TCO zram + compose + TCO

Docker Compose & Self-Hosted AI Stack RAM, Swap & VPS Sizing Calculator

Sum empirical container peak RSS across 12 services, evaluate 2 GB to 32 GB OOM risk (Exit Code 137), export zram-generator and hardened 127.0.0.1 Compose configs, and calculate Cloud VPS vs. local Mini PC break-even months.

COMPOSE V2 · UFW LOOPBACK docker-compose.yml + .env

Docker Run to Production Compose Converter

Parse multi-line docker run CLI commands into hardened Docker Compose v2 YAML and separated .env secret templates with automatic 127.0.0.1 loopback port guards, healthchecks, and log caps.

CATEGORY 03 // QUANTIZED WEIGHTS, GQA KV CACHE & UMA

Local LLM & Unified Memory Hardware

NEW · 36M CLOUD VS LOCAL AI TCO · BREAK-EVEN kWh + Egress + Resale SVG

Cloud VPS & GPU vs Local AI Hardware 36-Month TCO & Break-Even Calculator

Compare 36-month Cloud VPS, Managed Hosting, and hourly GPU bills (including bandwidth and snapshot egress) against owned local AI workstations (Mac mini M6 32GB, RTX 5070 Ti 16GB GDDR7, and NVIDIA GB10 128GB UMA) with wall-plug kWh power math and 3-year resale recovery.

GGUF · AWQ · GQA KV · GB10 128GB UMA Ollama + vLLM + llama-server

Local LLM VRAM, KV Cache & Decode Bandwidth Calculator

Compute exact Transformer quantized weights (FP16 to Q4_K_M and AWQ), Grouped-Query Attention KV cache growth across context lengths (2K-128K) and parallel slots, single-stream decode tok/s, and simultaneous fit verdicts across RTX 5070 Ti, RTX 4090, Mac mini M4, and NVIDIA GB10 128GB UMA.

CATEGORY 04 // RETINA PPI, WINDOWSERVER BACKBUFFER & DSC 1.2A

macOS External Display & Bandwidth

MACOS HIDPI · 5K2K · TB4 / TB5 DSC VESA CVT-RBv2 + BetterDisplay CLI

Mac External Monitor PPI, HiDPI VRAM & DP/TB Bandwidth Calculator

Calculate physical pixel density (110 vs 218 PPI Retina zones), macOS WindowServer 2x HiDPI backing framebuffer Unified Memory (7680x3240 for 3840x1620 scaled 5K2K), and VESA CVT-RBv2 link rates across DisplayPort 1.4, HDMI 2.1, Thunderbolt 4, and Thunderbolt 5.

OPEN-SOURCE DIAGNOSTIC CLI // MIT LICENSE Python 3.9+ Stdlib · 12 Checks < 1.0s · 60 Unit Tests · 100% Read-Only

Open-Source Host Auditor: gfl-doctor v1.5.0 (12 Read-Only Checks)

gfl-doctor v1.5.0 (bundled in the MIT-licensed gearflowlab-benchmarks open-source archive) audits Linux x86_64/ARM64 servers and Apple Silicon workstations across 12 read-only checks in under 1.0 second using strictly allowlisted probes and zero third-party packages. When a check flags a risk, it outputs a prefilled URL pointing straight to the matching interactive utility above.

1–2. Unlinked Logs (lsof +L1) & ARM64 QEMU Detects deleted-but-open Docker json-file descriptors and hidden linux/amd64 emulation on ARM64 hosts.
3–4. Wildcard 0.0.0.0 UFW & Nginx 60s SSE Flags sensitive ports bypassing Ubuntu UFW and audits Nginx proxy_buffering off on MCP/SSE routes.
5–6. PgBouncer 42P05 & macOS 2x HiDPI Verifies PostgreSQL/PgBouncer pools and Apple Silicon external monitor Retina backbuffer VRAM.
7–8. LLM KV-Cache UMA & Compose mem_limit Audits multi-slot FP16 vs Q8_0 KV-cache explosion, cgroup v2 memory caps, and Linux zram swap state.
9–10. Container Log Scan & Traefik acme.json Scans container tails for Exit 137 / 42P05 / 502 and checks 0600 ACME TLS & WebSocket headers.
11–12. 36M Cloud TCO Burn & Deadlock Watchdog Flags idle Cloud VPS/GPU spend via Utility #9 and unmonitored unless-stopped PID-1 hangs via Utility #10.
Quickstart & 60-Test Stdlib Suite (gearflowlab-benchmarks)
# 1. Run gfl-doctor v1.5.0 directly with Python 3.9+ stdlib (< 1.0s, 12 read-only checks)
python3 cli/gfl-doctor/gfl_doctor.py

# 2. Emit anonymized JSON diagnostics (including Checks #11 TCO & #12 Deadlock Watchdog)
python3 cli/gfl-doctor/gfl_doctor.py --json --anonymize

# 3. Run all 60 stdlib unit tests and verify SHA-256 benchmark archive integrity
python3 -m unittest discover -s cli/gfl-doctor/tests -v
shasum -a 256 -c SHA256SUMS.txt

Every test suite runs via the Python standard library unittest runner with zero third-party packages. Browse all verified server benchmarks in Engineering Guides & Playbooks.

OPEN-SOURCE AI AGENT MCP SERVER // MIT LICENSE MCP SDK >= 2.1.0 · 10 Tools · stdio + 127.0.0.1 Streamable HTTP · 64 KB Guard

Open-Source AI Agent MCP Server: gfl-devops-mcp v1.2.0 & /mcp.json

Give Claude Code, Cursor, Antigravity CLI (agy), and self-hosted n8n agents deterministic access to all 10 GearFlow Lab infrastructure engines across default offline stdio or loopback-hardened 127.0.0.1 Streamable HTTP mode (refusing wildcard 0.0.0.0 bindings unless explicitly unlocked). Built on the official Python MCPServer v2 SDK (mcp >= 2.1.0) with read-only tool annotations and a fast < 5ms 64 KB payload guard, every tool returns exact sizing math, production configs, and verified prefilled URLs. Machine-readable agent discovery is live at https://gearflowlab.com/mcp.json alongside /llms.txt and /llms-full.txt.

1. decode_devops_error_log Matches 62 Docker, Nginx, PostgreSQL, n8n, and GB10 signatures with cascade chain detection and Error Decoder links.
2. calculate_llm_vram_kv_cache Computes GGUF/AWQ weights, GQA KV cache GiB, decode tok/s, and 6-hardware fit matrices.
3. calculate_postgres_pgbouncer_pool Outputs RAM-safe shared_buffers, work_mem, and SQLSTATE 42P05 transaction pools.
4. calculate_docker_vps_ram_tco Models container peak RSS, Linux kernel/cache overhead, zstd ZRAM, and VPS RAM tier sizing.
5. generate_reverse_proxy_config Synthesizes unbuffered SSE/WS configs for Nginx, Caddy 2, Traefik v3, and Cloudflared via the Proxy Matrix Generator.
6. convert_docker_run_to_compose Converts docker run commands into UFW-safe 127.0.0.1 Docker Compose v2 YAML.
7. calculate_macos_hidpi_bandwidth Computes Retina PPI zones, 2x HiDPI Unified Memory MiB, and DP 1.4 / TB4 / TB5 DSC links.
8. run_local_gfl_doctor_summary Executes read-only anonymized gfl-doctor v1.5.0 (12 host checks) and returns prefilled remediation URLs.
9. calculate_cloud_vs_local_ai_tco Models 36-month kWh, resale, and exact break-even across Cloud VPS/GPU vs mac32, rtx5070ti, and gb10 via Utility #9.
10. generate_container_health_watchdog Generates Compose healthcheck + 2s grace, Linux systemd, and macOS LaunchAgent auto-healers via Utility #10.
1-Click Dual-Transport Setup (Claude Code, Cursor, Antigravity CLI & n8n)
Inspect /mcp.json →
# 1. Install gfl-devops-mcp v1.2.0 from gearflowlab-benchmarks (Python 3.10+, mcp >= 2.1.0)
pip install "mcp>=2.1.0" "uvicorn>=0.30.0" && pip install -e ./mcp/gfl-devops-mcp

# 2. Default stdio mode (Claude Code / Cursor / Antigravity CLI, 0 network sockets)
claude mcp add gfl-devops -- gfl-devops-mcp

# 3. Hardened 127.0.0.1 Loopback Streamable HTTP mode (n8n MCP Client / Local Daemons):
gfl-devops-mcp --transport streamable-http --host 127.0.0.1 --port 8765