Parallel All the Way Down: Beyond Single-Token Generation with Speculative Decoding
vLLM development signal - relevant to the current vLLM serving pin.
Independent publication // local-first AI // field reports from owned systems
Nodehome
A running publication about local models, private inference, self-hosted agents, weird hardware, research sweeps, and the builders wiring their own AI stack together.
AI is getting physical again. It shows up in terminals, racks, side projects, and ugly little workflows people actually control.
Latest
Daily Sweep - Jul 29, 2026A new TIL on adding custom MCP servers to both the ChatGPT and Claude regular chat interfaces - it's a little less obvious than I had hoped, but I got there in the end til.simonwillison.net/llms/mcp-in-...
Agent or tool-use pattern - relevant to local AI workflow automation.
datasette released 0.1a0 at datasette/datasette-agent-modeler
Simon Willison tool release - usually practical and worth evaluating.
simonw released 0.3 at simonw/mcp-explorer
Simon Willison tool release - usually practical and worth evaluating.
Anatomy of a Frontier Lab Agent Intrusion: A Technical Timeline of the July 2026 Incident
Simon Willison signal - surfaces practical patterns before they spread.
conversion: fix Qwen2.5-Omni mmproj conversion regression (#26262)
llama.cpp commit - read only if it touches your serving path.
simonw pushed til
Simon Willison signal - surfaces practical patterns before they spread.
Simon Willison: pushed datasette-agent-modeler (4 events)
Routine GitHub activity compressed for scanability.
Field Reports
builds, experiments, notesThree RTX 3090s, One 32B Model: A Pipeline-Parallel Canary
A current field note on why the 3x3090 serving path moved through pipeline parallelism, not tensor parallelism, for the tested 32B AWQ model.
Gemma 4 12B And The Sensory Agent Lane
A public-safe read on Gemma 4 12B as a local sensory preprocessor: useful for seeing, hearing, and structuring observations without turning into an action system.
Hardware
machines, thermals, economicsPower Caps On Three RTX 3090s: Bursts Versus Sustained Load
A measured note on 300W bursty inference, lower caps for sustained runs, and why power-cap sweet spots are workload-specific.
Parallel Agent Serving Is A Hardware Shape Now
A field-report read on 14x RTX 3090 agent serving, EXL3, FP8 KV cache, Aphrodite, and why concurrency is becoming the local hardware metric.