gpt-oss:120b generates its answer AFTER the reasoning pass. With
think=false AND a JSON format set it returns EMPTY content and no
thinking at all. Karakeep sends exactly that, so every tagging job
failed in under a second with 'the model ignored our prompt and
didn't respond with the expected JSON: {}'.
Verified by curl through the router:
think=false -> content ''
think=true -> content '{"tags":["x"]}'
Fix: keep reasoning ON upstream and strip the 'thinking' field from
the reply instead, which is what a caller asking for no thinking
actually wants. Applies to the native forge path (which forwards the
body as-is) and the cloud mapping. Streaming is untouched: HA streams
and already handles the separate field.
24 asserts forge's runner is permanently resident, not merely loaded at the
moment of the check — a cold load is 21-66 s and lands in front of the caller,
so "loaded right now" is not the property that matters.
25 is the regression that would otherwise be silent: an explicit finite
keep_alive from a caller BEATS forge's OLLAMA_KEEP_ALIVE default, so the router
override is the only thing stopping one such caller evicting the 64 GB model for
everyone else. The check sends keep_alive:"30s" natively and asserts residency
survives it.
Both pass on lounge and nidavellir, 25/25.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Ollama keeps one runner config per model and reloads the 64 GB model on
every request whose num_ctx differs from the loaded runner's. The router
now sets options.num_ctx = FORGEWAY_NUM_CTX (default 131072) on the Ollama
dialect and carries the same value in the probe (native /api/chat; the
OpenAI /v1 endpoint ignores options, verified). /status exposes the pin and
the observed runner context and warns on mismatch; check 23 asserts it.
Verified by behaviour: an 8192 caller, a 4096 caller, probes and 40k/100k-
token retrieval prompts in one window -> 0 reloads, 0 timeouts, HA turns
0.3-0.6 s; before, 3 rounds of an 8192 caller = 7 loads and two probe
timeouts. 23/23 acceptance on both clusters.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Two Ollama entries — "Local only — private (fails if forge is down)"
(local-strict) and "Local preferred — falls back to OpenAI cloud"
(local-optimistic) — created/renamed headlessly with scripts/ha_ollama.py
(REST config flows + WebSocket renames; wait for the entry to be loaded
before opening a subentry flow; entities matched by subentry id).
Each proven by a real conversation through HA, and the difference proven
through HA against a throwaway dead-upstream instance published under a
temporary .services. name: strict → HA error, no cloud; optimistic → cloud
answer in ~1 s. All temporaries removed. Rick's pfSense rule (HA IP →
ingress) verified from HA's side; alias is NXDOMAIN publicly, so a loaded
entry proves internal resolution.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
/{mode}/api/{tags,chat,generate,version,show,ps} beside the OpenAI surface.
forge path is a raw proxy (forge is Ollama); the cloud fallback is translated
both ways — NDJSON frames byte-compatible with Ollama's, tool calls round-trip
with synthesised ids, format mapped, think→reasoning_effort (forced to none
with tools: OpenAI rejects the combination on gpt-5.4-mini). Tags never wait
on a model load (HA's list() has a 5 s timeout). local-strict still 503s in
Ollama's error shape; the OpenAI dialect is untouched.
Home Assistant (Lounge, 2026.9.2) configured through its REST config flow and
proven with a real conversation. HA's VLAN cannot reach the k8s ingress, so
it uses forgeway.dmz.lounge.falkvinge.net through the entrance HAProxy
(RFC1918-fenced vhost); the .services. alias needs a pfSense rule from Rick.
22/22 in-cluster checks on both sites. scripts/ha_setup_ollama.py is
idempotent and ends with a real answer; local-smoke-ollama.sh uses the exact
client HA pins.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Rick confirmed forgeway.<site>.falkvinge.net and the local-first alias
forgeway.services.falkvinge.net. README documents the alias as the front
door, the per-site names as explicit handles, and exactly how the Serenity
override keeps the alias LOCAL (host-record + local= on both Serenity
Pi-holes; the local= line closes the dual-stack AAAA bypass).
OpenClaw providers now point at the https alias with a placeholder apiKey:
OpenClaw exempts plain-IP http base URLs from needing a key but not https
hostnames. scripts/technitium.py ships the Technitium API helper used for
the records (zones / records / add / resync / logtail).
REPORT.md addendum carries the verification table from real clients at both
sites and the findings made on the way.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Two endpoints over forge's Ollama gpt-oss:120b:
/local-optimistic/v1 forge, else cloud gpt-5.4-mini
/local-strict/v1 forge, else a clean 503 the caller handles
Discovery, not configuration: each instance probes an ordered candidate list
(stable DNS name, then the Lounge and future Serenity literals) concurrently but
in priority order, so the happy path is one round trip and silicon's move to
Serenity needs no rewrite.
/healthz runs a real 1-token completion — forge can be up with Ollama stopped.
Liveness/readiness are deliberately NOT forge-gated: otherwise k8s would evict
the pods during an outage and callers would get a connection failure instead of
the promised 503, and Kuma could not tell "forge down" from "router gone".
max_tokens/max_completion_tokens are translated per backend: Ollama accepts only
the former, gpt-5.4-mini rejects it outright.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>