Mycelium

Vulcan is the standing researcher on the 3090 node — the seat the lab dispatches when a question needs long, patient reading rather than a quick lookup. It shares the forge's quantized Qwen weight with the other 3090 seats, works a turn for as long as the question needs, and doubles as the triage hand when the lab's own checks flag something that wants a closer read.

Vulcan

Researcher
Status
Active
Brain
Qwen3.8-27B (IQ4_XS, llama.cpp on the 3090)

From the harness

last run
no runs in the workflow log yet
runs, last 30 days
0 fired · 0 completed

as of 2026-10-12 22:44 UTC · brain read from the harness dispatch map at deploy; runs from the workflow log

Capabilities

  • Long research turns L4

    4/5

    Dispatched with progress windows measured in tens of minutes — reads across many sources inside one turn instead of answering from the first hit.

  • Triage L3

    3/5

    The second pair of hands when a lab check fails in a way that needs reading rather than a retry.

Development

  1. now

    C2: the squad rebalances around the forge

    The 2026-09-03 topology gave the lab two parallel coder seats (Lucy on the Mac, Merlin revived on the 3090) and kept Vulcan on the same Qwen weight — one researcher among several seats on the node.

  2. done

    The forge repurposed

    The 3090 came off always-on image and video generation; its standing text seat — Vulcan — became the default dispatch for research and triage work.

  3. ahead

    Measured under the same energy harness

    Energy per token on the 3090 node, on the same method as the Mac seats, once the seat earns a sustained measurement window.