From 1dac2abeb29c8779a9f3a8fd84f4411ad1ba986e Mon Sep 17 00:00:00 2001
From: ValueOn AG
Date: Sun, 10 May 2026 22:10:00 +0200
Subject: [PATCH] abo enterprise, ai agent fixes
---
c-work/_CHANGELOG.md | 12 ++++++++++++
.../1-analyse/spezifikation.txt | 16 ++++++++++++++++
2 files changed, 28 insertions(+)
create mode 100644 f-decisions/infra-gpu-instance/1-analyse/spezifikation.txt
diff --git a/c-work/_CHANGELOG.md b/c-work/_CHANGELOG.md
index c1e4183..bb42edd 100644
--- a/c-work/_CHANGELOG.md
+++ b/c-work/_CHANGELOG.md
@@ -12,8 +12,15 @@ type: `feat` `fix` `refactor` `docs` `test` `chore` `build` · scope: `gateway
Skip: reine Refactors, Formatting, Lint, Dep-Bumps, Test-only, Wiki-Tippfehler.
+## 2026-05-10
+
+- 2026-05-10 | fix | gateway | CommCoach: conversational turns use `operationType=DATA_QUERY` + `priority=SPEED` for faster model selection (gpt-5.4-mini instead of gpt-5.5); model selector speed priority weight scaled from `/10` to `*100` to be meaningful.
+- 2026-05-10 | fix | gateway | CommCoach: persona sessions now distinguish first vs follow-up; opening prompt references prior sessions instead of re-introducing persona every time.
+- 2026-05-10 | feat | gateway | Document generation: unified `documentTitle` style separate from `headings.h1`; BASE/renderer converts independently; HTML/CSS, PDF title block, Markdown/Text heading offsets updated; DEFAULT_STYLE h1 lowered vs title.
+
## 2026-05-09
+- 2026-05-09 | feat | actan | SanctionsCheck: Login rememberMe + session cookies; public `/signup` + `POST /api/public/signup`; e-mail templates + `templateRenderer`; SMS retries + NOTIFICATION activities; profile tabs; activity API enrichments (`akteurTyp`, `sichtbarkeit`, `includeInternal`) and Activities UI badges/panel.
- 2026-05-09 | feat | gateway | Enterprise Subscription: sysadmin-managed flat-price subscriptions with custom limits (users, features, storage, AI budget), invoice email, hard-block on overage, auto-renewal cron job. New endpoints: POST enterprise/create, enterprise/renew, PUT enterprise/update. (c-work: 1-plan/2026-05-enterprise-subscription.md)
- 2026-05-09 | refactor | frontend-nyla, gateway | CommCoach: remove Dossier view (route, FeatureView registry, RBAC UI object, Dashboard link, pageRegistry icon, barrel export). Session details (summary, score, duration, export) moved into Modules expand view.
@@ -231,4 +238,9 @@ Skip: reine Refactors, Formatting, Lint, Dep-Bumps, Test-only, Wiki-Tippfehler.
- 2026-05-07 | feat | gateway | RedmineTicketMirror + RedmineTicketDto: Feld `doneRatio` (% erledigt) ergaenzt; Sync-Mapping und DTO-Konvertierung aktualisiert
- 2026-05-07 | feat | gateway | Neues Agent-Tool `redmine.listRelations` fuer direkte Abfrage der Beziehungstabelle (Filter: issueId, relationType, pagination)
- 2026-05-07 | feat | gateway | `redmine.listTickets`: `offset`-Parameter ergaenzt fuer echte Pagination; Response liefert `offset` + `hasMore` statt `truncated`
+- 2026-05-10 | feat | gateway | CommCoach Fast Conversational Mode: `AgentConfig.excludeAllTools` flag fuer tool-freie Turns; generischer RAG-Session-Cache in `agentLoop` (TTL 120s / 5 Msgs); CommCoach nutzt beides fuer fluessige Coaching-Gespraeche
+- 2026-05-10 | fix | gateway | RAG-Cache-Bug: doppelter `workflowId`-Parameter in `_getOrRefreshRag` behoben (Keyword-Kollision)
+- 2026-05-10 | fix | gateway | CommCoach TTS: `_TTS_WORD_LIMIT` von 200 auf 80 Woerter reduziert; Sprachausgabe fasst laengere Antworten jetzt zusammen statt alles vorzulesen
+- 2026-05-10 | feat | gateway | `AgentConfig.priority` (speed/quality/cost/balanced) an `agentLoop` durchgereicht; CommCoach conversational turns nutzen `PriorityEnum.SPEED` fuer schnellere Modellauswahl
+- 2026-05-10 | fix | gateway | `interfaceDbManagement.getInterface` Singleton Race-Condition behoben: shared mutable instance ueberschrieb featureInstanceId zwischen async Requests; Agent konnte Files nicht finden
diff --git a/f-decisions/infra-gpu-instance/1-analyse/spezifikation.txt b/f-decisions/infra-gpu-instance/1-analyse/spezifikation.txt
new file mode 100644
index 0000000..88da7e4
--- /dev/null
+++ b/f-decisions/infra-gpu-instance/1-analyse/spezifikation.txt
@@ -0,0 +1,16 @@
+# SPEZIFIKATION
+Private LLM GPU Instance
+
+## SERVER
+1× Server Supermicro AS-4125GS-TNRT (4U Chassis mit Netzteilen, Mainboard, Kühlung)
+1× CPU AMD EPYC 9354 (32 Cores)
+12× 32 GB DDR5 ECC RDIMM (= 384 GB RAM)
+2× SSD Samsung PM9A3 1.92 TB NVMe (für das Betriebssystem)
+2× SSD Samsung PM9A3 3.84 TB NVMe (für Modelle und Daten)
+1× Netzwerkkarte Mellanox ConnectX-6 Dx 2× 25 GbE + 2 passende Transceiver oder DAC-Kabel
+1× Rail-Kit für den Server (passend zum Chassis)
+1× USV APC Smart-UPS SRT 3000 (3 kVA)
+1× Rack-Steckdosenleiste (PDU) mit C13/C19-Anschlüssen
+
+## PRO GPU
+1× GPU NVIDIA L40S 48 GB