Systems
Reading live sources…
Each page here reads the same APIs the operator uses, sanitized, and every panel says whether its figures are live, a snapshot, or unavailable.
Figures are read when this page renders. A dash means the source did not answer; nothing is filled in.
Mills, the agent pipeline that plans, implements, tests, and merges changes to the published repositories. Every run is on the record, including the ones the gates refused.
merged / runs, 30 days
510 escalated · $6.61 per merge
LiveOpen models served by FlexInfer on self-hosted GPUs: which are loaded now, on which engine and hardware, and how fast they decode.
running / configured
Engines: diffusers · llama.cpp · vLLM
LiveThe K3s cluster the site and its models run on: nodes by role, capacity, and a health score, with names replaced by stable tokens.
nodes ready / total
4 GPU nodes · 4 GPUs · 622 pods · health 98/100
LiveThe Go-native MCP servers Loom Core exposes to coding agents, each with the tools an agent may call without asking.
MCP servers
498 allowlisted tools · 86 category tags
Snapshotedilint checks X12, HL7 v2, EDIFACT, and delimited files for the defects a trading partner would reject. It is compiled from Go to WebAssembly.
runs in your tab
Nothing you paste or open is uploaded.
LocalEvery figure comes from the site’s public API. Node and pod names are replaced with stable tokens, ids are hashed, and factory runs carry only allowlisted fields. Read the JSON: factory · models · topology · metrics.
These were the site’s demos before the systems above had public APIs. They still load at their original addresses; each tag says what data the demo shows.
Plays a repository’s Git history back as audio. The app is hosted separately at git-tunes.flexinfer.ai.
separate appNode health, pod spread, and a health score from the FlexDeck topology API. The cluster view replaces it.
live FlexDeck dataA stage-by-stage pipeline timeline. The pipelines shown are sample data, not this site’s CI.
Model cards from the FlexDeck model API, with a toggle for sample models. The inference plane replaces it.
live, sample toggleA canvas animation of the news-analyzer stages: scrape, extract, summarize, deliver. The flow is simulated.
simulatedTraffic from Cloudflare through ingress to services, with the GitOps path. A fixed diagram, not read from the cluster.
static diagramA scroll story of the Dell R730xd host, its Harvester VMs, and the K3s nodes they form.
March 2026 snapshot© 2026 FlexInfer.