https://gitlab.synchro.net/main/sbbs/-/commit/f5f55249bd04c9133041aee7
Modified Files:
ctrl/chat_llm.ini exec/chat_llm_irc.js
Log Message:
chat_llm: fix stock chat_llm.ini doc drift + wire opening system prompt

Cleanups to the config shipped in 39502cc1f:

- Wire opening_system_prompt_file = chat_llm_opening_persona.utf8 so the
bundled opening-turn persona is actually used; it shipped but was never
referenced, so the opening greeting fell back to the full system prompt.
Matches the working live config.
- Drop textdir/sourcecode/gitlab/git_log from the index_sources comment;
only msgbase/filebase/dokuwiki crawlers ship in exec/llm_index/.
- Replace the stale "v0 sketch -- no RAG, no persistent memory yet" header;
both RAG and persistent memory now exist.
- chat_llm_irc.js: header/inline comments said irc_channel; the code reads
irc_channels (comma-separated list of channels).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

---
þ Synchronet þ Vertrauen þ Home of Synchronet þ [vert/cvs/bbs].synchro.net

https://gitlab.synchro.net/main/sbbs/-/commit/3ef83bcbfc9ed41607b63774
Modified Files:
ctrl/chat_llm.ini exec/chat_llm.js
Log Message:
chat_llm: add configurable num_ctx (Ollama context window)

Adds a num_ctx knob (read from chat_llm.ini, per-persona or [default])
passed through to Ollama's request options on all three /api/chat
builds (initial, pre-tool, follow-up). 0/unset => omitted via
`cfg.num_ctx || undefined` (JSON.stringify drops an undefined value),
so Ollama keeps its server default and existing installs are unaffected.

Why: a large system prompt plus retrieved RAG context can exceed
Ollama's ~4096-token default; llama.cpp then keeps the prompt's tail and truncates the FRONT, dropping the identity/style rules at the top of the
system prompt and making the bot answer as a generic "AI assistant".
Observed with a 4698-token persona prompt capped at 4096. Documents the
knob (commented, off by default) in the stock chat_llm.ini.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

---
þ Synchronet þ Vertrauen þ Home of Synchronet þ [vert/cvs/bbs].synchro.net

https://gitlab.synchro.net/main/sbbs/-/commit/143338a4deabfd87458fee78
Modified Files:
ctrl/chat_llm.ini exec/chat_llm.js
Log Message:
chat_llm: endpoint failover (endpoint_fallback)

Add an optional endpoint_fallback. dispatch() now wraps the provider
call: when the primary endpoint throws (transport error / non-200), it
retries once against the fallback, and a ~60s cooldown routes straight
to the fallback on subsequent turns so an outage doesn't add the
primary's full timeout to every reply (re-probes the primary after the cooldown). No fallback configured => unchanged behavior. Documents the
knob (commented) in the stock chat_llm.ini.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

---
þ Synchronet þ Vertrauen þ Home of Synchronet þ [vert/cvs/bbs].synchro.net

https://gitlab.synchro.net/main/sbbs/-/commit/f4630b0a17237b8f8ef08728
Modified Files:
ctrl/chat_llm.ini exec/chat_llm.js chat_llm_irc.js llm_index.js exec/llm_tools/relay_message.js
Log Message:
chat_llm: per-name data/chat subdirectory layout

Move per-persona chat_llm files out of the flat data/chat/ namespace into per-guru subdirectories data/chat/<name>/ (name = persona code before ':'), eliminating filename-prefix collisions and the safe_id ':'<->'_' ambiguity (guru:irc vs guru_irc) the flat scheme allowed.

data/chat/<name>/ holds:
- llm_rag.idx RAG index, shared across the guru's modes
(IRC bot + terminal guru)
- irc_relay.json, irc_mute.json, irc_norelay.json, irc_seen.json
IRC state, proto-prefixed
- <speaker>[.<proto>].json per-speaker memory; the proto tag keeps a
speaker's terminal vs IRC memory from colliding Naming: '_' joins fixed name-parts, '.' separates the variable proto/ext.

- chat_llm.js: persona_parts()/persona_dir() helpers; memory_path() ->
data/chat/<name>/<speaker>[.<proto>].json; index_output default ->
chat/<persona>/llm_rag.idx. User-numbered speaker ids zero-padded to
%04u (user_0001) to match Synchronet's data/user/0001.* convention.
- llm_index.js: writes chat/<name>/llm_rag.idx, keyed on the NAME part so
`llm_index.js guru` and `... guru:irc` agree with load_index().
- chat_llm_irc.js: PERSONA_NAME/PROTO/DIR; IRC state under PERSONA_DIR;
chat log + .announce stay in data root (short hand-touch paths).
- relay_message.js: off-session defaults under data/chat/guru/; the
recipient history-scan now reads the persona dir (from env.relay_path).
- ctrl/chat_llm.ini: index_output = chat/<persona>/llm_rag.idx.

Live data migrated into the new layout (33 files; index moved by rename,
no rebuild). Validated: path resolution, full regression unit suite, and end-to-end relay opt-out honored from the new path.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

---
þ Synchronet þ Vertrauen þ Home of Synchronet þ [vert/cvs/bbs].synchro.net

https://gitlab.synchro.net/main/sbbs/-/commit/348e8ae9f0468c124b827d82
Modified Files:
ctrl/chat_llm.ini exec/chat_llm.js chat_llm_irc.js chat_llm_routes.js exec/llm_tools/relay_message.js
Log Message:
chat_llm: faithful relays -- verbatim routing, perspective rewrite, join delivery

Rework message-relay so a relayed line reaches the recipient as intended
rather than paraphrased or pronoun-inverted:

- Broaden the deterministic relay router (chat_llm_routes.js) to catch many
more phrasings -- please/can-you preambles, and ask / remind / leave a
message for / send / give / message / say-hi-to / if-you-see / when-X-is-
back -- so the body is stored VERBATIM instead of falling through to the
model, which rewrote it (3rd person, added preamble, once a whole poem).
Pronoun/article guards keep "let me know", "say hi to everyone", etc. from
false-matching as relays.

- Perspective rewrite: a relay phrased from the sender's seat ("tell Dan you
love his work") is re-anchored to the recipient's point of view ("I love
your work") before it's stored. Gated on pronoun presence -- pronoun-free
bodies ("the build is fixed") stay byte-verbatim with NO model call.
relay_rewrite_model points this small, infrequent call at a stronger model
(subtle reference task a 7B does unreliably), kept transient via keep_alive
0 so the faster chat model stays the resident, warm one on a host that
can't hold both in VRAM.

- Deliver pending relays on JOIN too, not just on speech: a relay request
says "next time you SEE X", and re-joining counts. deliver_pending() drains
the queue, so this is idempotent with the speak-triggered path.

- Delivery line quotes the relayed body; acknowledgement wording is now
"joins or speaks". Bot log timestamps switch to local-time ISO-8601 via
strftime() instead of UTC Date.toISOString().

New ctrl/chat_llm.ini keys (documented in the stock file): relay_rewrite, relay_rewrite_model, relay_rewrite_keep_alive.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

---
þ Synchronet þ Vertrauen þ Home of Synchronet þ [vert/cvs/bbs].synchro.net

https://gitlab.synchro.net/main/sbbs/-/commit/135f08f0f5e1146d28b3c9ed
Modified Files:
ctrl/chat_llm.ini exec/chat_llm.js chat_llm_irc.js
Log Message:
chat_llm: stop the IRC guru volunteering wrong/unsolicited answers

Two changes to make the guru a far more reluctant, more accurate
unprompted participant on IRC, prompted by live #synchronet logs where
it (a) answered questions one user had explicitly aimed at another, and
(b) confidently restated others' mistakes as fact.

1. directed_at_other() (chat_llm_irc.js): a question addressed to a
specific other participant ("nelgin: why are you changing ip?") is a
person-to-person exchange, not a room question -- the bot no longer
arms an intervention on it. is_nick_ping only caught a BARE "Nick?";
this catches "Nick: <content>" / "Nick, <content>" where Nick is a
seen channel member that isn't the bot.

2. Two-layer confidence gate for unprompted interventions only (direct
questions are answered best-effort as before):
- In-prompt SKIP escape: the volunteering generation is told to answer
only if the retrieved docs explicitly support a correct answer and
to never restate an asker's assumption as fact; otherwise it emits
SKIP and the bot stays silent (and persists nothing). No extra
round-trip. (ctx.volunteering -> build_messages; is_volunteer_abstain
nulls the reply in chat_session.)
- 14B fact-check (verify_volunteer_answer): a 7B feels "grounded" off a
merely-related chunk and confirms false premises anyway (observed: a
bogus mods/exec load path off a chunk about mods shadowing exec). Any
non-declined answer is fact-checked by a stronger model that must
return VERIFIED; fails closed (any error/ambiguity -> silence). Reuses
relay_rewrite_model, kept transient (keep_alive 0) so the warm chat
model isn't evicted. New ini knobs: intervention_verify[_model|_prompt].

Also adds an anti-sycophancy clause to the IRC grounding for all replies:
don't agree with or repeat a user's claim unless the docs support it.

Live-validated: the false-premise "mods/exec" question now abstains
(3/3), an ungroundable question abstains, and legitimate groundable
questions (SBBSecho, transfer protocols, BinkP) still answer.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

---
þ Synchronet þ Vertrauen þ Home of Synchronet þ [vert/cvs/bbs].synchro.net