vv_is_ai_host() was `=== 'host1'`, which made a physical fact — that is where the
GPU is — look like a rule. AI_OWNER_HOST declares it, so the card can move to a
rebuilt host3 or a friend's spare.
The gate was also doing two jobs. Assistant docks and findings strips now ask
whether a model is reachable, so a node without a GPU gets them by borrowing;
the AI tab asks whether this is the owner, because that page carries the bug
reports, the index and the model configuration — the surface where the
vocabulary assumes you built the mesh.
Resolution is local, then owner, then anyone else declaring a model, pinned once
it answers. Pinned rather than re-derived per call: a mesh that re-decides every
request eventually decides differently mid-conversation, and a chat whose second
turn lands on another machine has no history there. Cleared only on a transport
failure, and only when there is somewhere else to go — a single-node mesh whose
model is down should say so, not report AI as unconfigured.
A stored transcript is replayed into a later prompt when reopened, so a key typed while
changing a setting would be handed back to the model on every subsequent turn.
Several headers argued the Scheduler dock was deliberately separate, and the Scheduler's help
never mentioned the assistant at all — including the fix flow that just changed shape.
"What is Varaverk" matched the PURPOSE intent, so retrieval returned every
script's one-line purpose and the model answered that the context does not
define the system — while README.md sat in the index unread. A definitional
question with no explicit filter now goes to kind=readme.
Varaverk is a coined word with no spell-check, so it arrives as varavrk,
veraverk, varavek. Edit distance catches those without a pattern that needs
extending per typo. This is search, not identity — a near-miss only widens a
document search, unlike hostname resolution where it must never resolve.
A small operator-written file handed to the model at the start of every
conversation — who you are, how this install is set up, what has already been
decided. Injected ahead of the retrieved passages and marked operator-authored
so it outranks anything they contradict, and never cited as a source.
Deliberately not indexed and deliberately under DATA_DIR: it changes
constantly, vector similarity is the wrong way to retrieve things you were
told to remember, and gitignoring it keeps personal notes out of a pushed
repository. The character cap is a context budget — this text costs its share
of 16k on every single turn.
vv_meta uses columns k/v, so the embedder-match check read nothing and
reported "unrecorded" — a silent pass on the one mismatch that returns
confident nonsense instead of an error. And Ollama reports tags fully
qualified while conf carries the bare name, so an installed embedding model
was reported missing. A health panel that cries wolf is worse than none.
The failures this subsystem actually has are configuration drift, so each
check names the setting to change rather than reporting that retrieval
failed. Notably it catches a conf model tag that is no longer installed, and
an index built by a different embedder than the one configured — vectors
from two models are not comparable, and that failure returns confident
nonsense rather than erroring. Diagnostic questions also get recent log
warnings, attached only then because they cost budget the passages need.
Token and poll rather than SSE, so the api layer keeps one response
convention and reuses the pattern manual_sync already proved. History is
capped at three turns because the model is only fully offloaded at 16384
context and unbounded history would cross that silently. The tab exists
only while AI_ENABLED is true, rejected server-side and not merely hidden.
Gmer4Lfe