Field notes
Specific deployment stories from local-AI builds we've actually run. Not buyer guides. Not "best GPU for X" lists. Closer to a working notebook — what we set up, what surprised us, what we'd change if we did it again.
Most local-AI hardware coverage is purchase-decision content — which is fine and useful, and most of this site is that. But the editorial work that compounds slowest and longest isn't ranked picks; it's specific stories about specific deployments. What broke at month four. What the spec sheet said vs what actually happened on a hot Wednesday at 4pm. The retailer note we were too embarrassed to write at the time.
These pages don't have ranked picks at the bottom. Some of them recommend something; most don't. They exist for the same reason a working engineer writes a postmortem: because the specific story is the part that helps the next person who's about to do the same thing.
Six months on a used RTX 3090: what actually broke
Bought a used 3090 in November 2025 for $820. Six months of daily inference later, here's what failed, what didn't, and the maintenance call I'd make differently if I were buying again.
Dual 3090 vs single 5090 — the deployment we actually built
Spent three weekends comparing the two builds for a 70B-class inference workload. Here's the rig we shipped, the one we rejected, and the spec-sheet difference that turned out to matter least.
New field notes get added when we finish a deployment worth writing about. No publishing schedule, no SEO-driven cadence.
Why field notes exist
Most local-AI content is theoretical. Tutorials walk through a happy path; benchmark tables list numbers; buyer guides rank hardware in the abstract. None of that survives the moment a real deployment hits week three and starts accumulating boring problems — the cooling profile drifts, a driver auto-update breaks vLLM, an SSD fills with embeddings nobody documented, the budget wasn't for the PSU upgrade that turned out to be necessary, the “great deal” used GPU has 4 GB of dead VRAM the seller didn't mention.
Field notes are the post-mortems. Each one covers a real deployment after at least a few weeks of actual use — the unanticipated friction, the workarounds that survived, the parts of the original plan we abandoned, the receipts for what we spent vs what we budgeted. The point isn't to score the hardware; it's to give the next operator a realistic preview of what their version of this deployment will feel like by month three.
How they fit alongside the buyer guides
Buyer guides answer “what should I buy?” Field notes answer “what was it actually like to live with what I bought?” A buyer guide that says “the used 3090 is the value pick at 24 GB” is true; a field note that says “six months later, my used 3090 is at 75°C ambient, on its second set of fans, and worth every dollar” is the trust signal that makes the buyer guide's recommendation believable.
If you've done a deployment that wasn't covered here and want to contribute, the contact page is the right place to start. Field notes are bylined; they carry your name, not ours.