1732 posts tagged with "Self-Hosting"
Running your own PaaS and infrastructure on machines you own
One API Key vs 7,000 Tools: Where a Self-Hosted PaaS's MCP Server Should Land
A homelab MCP bridge runs your infrastructure behind one API token; Arcade's engine vaults per-user OAuth behind 7,000 tools. The head-to-head operator math at three team sizes, and where a self-hosted PaaS should land between them.
Dedicated vs VPS for AI Agents: 64 GB of Exclusive RAM for $57/Month Changes the Hosting Math
osModa's 2026 comparison prices a Hetzner AX42 dedicated box with 64 GB of exclusive RAM at ~$57/month against a 16 GB shared VPS at ~$80. This post reproduces the $/GB math, works out the bin-packing crossover for fleet operators, and shows why the dedicated-vs-VPS line lands differently for agents than for web services.
Deplo Takes Your VPS Over: What a Migration-First Panel's 66-of-67 Dokploy Import Really Proves About Single-Box Lock-In
Deplo's installer reads a Dokploy or Coolify panel over its own API and recreates it with data included — 66 of 67 apps landed untouched in its September 4 audit. The named misses reveal where single-box lock-in really lives, and why the move still stops at per-app placement.
Django Hosting in 2026: $140+/Month on Heroku vs $38+ on Render vs $15–35 on Fly.io — Migrate the Bill or Exit Metered Compute?
The same production Django stack costs $140+/month on Heroku, $38+ on Render, $15–35 on Fly.io, or a flat $5–40 on owned hardware. A like-for-like price breakdown plus the Django migration checklist for each path.
Dokku vs Dokploy vs Kubero: Three Generations of Self-Hosted Heroku Alternatives
Dokku, Dokploy, and Kubero are three generations of the same idea: Heroku-style deploys on infrastructure you own. A head-to-head on deploy flow, scaling, HA, multi-node reality, and upgrade pain, plus where declarative machine provisioning fits next.
Dokploy in 2026: How Close Is Single-Server PaaS to Cluster API?
Dokploy spent 2026 adding remote servers, previews, external secrets, and agent tooling. This inventory sorts which additions actually raise the single-box ceiling from dashboard polish, and names the machine-layer line that still needs declarative provisioning.
Stop Letting GPU Requests Pend: Kubernetes 1.36's DRA Prioritized Alternatives
Exact-match GPU requests leave pods pending while other card types sit idle. Kubernetes 1.36's stable DRA prioritized lists let one claim say H100, A100, or T4 — here is the claim-template design, the scoring behavior, and four gotchas for mixed-card fleets.
Envoy AI Gateway Hits 1.0: What an LLM-Aware Kubernetes Gateway Actually Load-Balances On
Envoy AI Gateway reached 1.0 as the standard routing layer for self-hosted LLM traffic on Kubernetes. How InferencePool endpoint selection on KV-cache state and queue depth beats round-robin — with benchmarks — and what it means for fleets you run yourself.
One Config Field, 12,000 Exposed Boxes: What Flowise's CustomMCP RCE Teaches Anyone Exposing Agent Tools
Flowise's CustomMCP node turned a config string into remote code execution on 12,000+ internet-exposed instances. The anatomy of CVE-2025-59528, the sibling CVEs proving it is a pattern, and a six-rule checklist for anyone running an MCP or agent-tool surface.