Field notes from our build queue.
Short posts on what we are seeing across self-hosted AI deployments. Practical, opinionated, written by the engineers who ship the stacks.
Why teams are self-hosting AI in 2026
For most internal use cases, self-hosting open-source models has become a credible default. Here is why it makes sense — and where it still...
Choosing the right open-source model for your server
A no-nonsense sizing guide. RAM, CPU, GPU, quantisation — what actually matters when you deploy on hardware you own.
RAG without the hype: what actually works
After twelve production RAG deployments, three patterns hold up. Three more that everyone tries don't. Here is the field notes version.
n8n vs Temporal vs Airflow for AI pipelines
Three workflow engines, three very different sweet spots. We pick n8n for most teams — here is when we don't.
How we secure self-hosted AI stacks
A walkthrough of the unglamorous-but-essential configuration that every stack we ship gets — TLS, auth, fail2ban, allowlisting, credential r...