- SAIA rate limits: minute=30/1m active at provider level; hour/day rows exist but not linked
- VK governance: allow_all_keys SQL workaround, allowed_models must be non-null JSON text
- Fix stale sk-tapirnase reference in add-MCP-client example (now uses cookie auth)
- Add LLM provider table
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- mcp-gateway.md: full rewrite for Bifrost at ai.nuclide.systems/mcp;
29 clients / ~760 tools; VK auth; n8n+shepard blocked on
streamable-HTTP; legacy gateway decommission checklist added
- connection-hosts.md: Bifrost at port 14003 is now ai.nuclide.systems;
LiteLLM demoted to internal-only; mcp.nuclide.systems marked pending
decommission; OIDC client 78c78998 flagged for removal
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Per-host + per-service LAN-only aliases added 2026-05-24 on CT 102.
Stable names → host IPs; clients use names instead of bare IPs so
service moves only require an AdGuard edit, not N app updates.
Use case driver: Immich Android HTTP/2 Keep-Alive bug — per-SSID
'nuclide' local URL https://immich.nuclide.lan:12000 bypasses Zoraxy.
Covers what Nexa is, which existing services it depends on, what is
genuinely missing (TEI, Qdrant collection, GraphDB), and alternative
tool choices for embeddings, web-search, and the graph pillar.
Repo: https://git.nuclide.systems/fkrebs/nexa
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Client IDs created programmatically via Pocket-ID /api/oidc/clients endpoint.
Env vars injected into both compose files and force-recreated.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
integration_openai points to LiteLLM at ai.nuclide.systems/v1.
Fixed completion=llama-3.3-70b-instruct, T2I=saia-flux (was devstral).
occ snippet for future model changes.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- New /ui/models page (sortable model catalog + service assignments)
- model_assignments.json: 35-model catalog with latency/cost/capability metadata
- API: GET/PATCH /api/model-assignments
- Model evaluator agent added to agents.json
- ui.html: widen max-width to 1600px, 4K breakpoints at 1920/2560/3840px
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- infra/connection-hosts.md: every host, LAN IP, port, and public URL
across all Proxmox guests, Docker services, network infra, and
external hardware; OIDC client registry; SSH cheat-sheet
- services/mcp-servers.md: all 27 MCP servers with full credentials,
upstream URLs, transport, group, gateway management API examples,
and gateway .env token inventory
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Add RESUME.md for cross-session continuity (open items, key state, constraints)
- claude-max-bridge: implement /v1/responses (OpenAI Responses API) with
previous_response_id chaining via server-side history injection into system prompt.
Root cause of "session already in use": CLI leaves JSONL in un-resumable
"dequeued" state after each --print run; fix avoids session reuse entirely.
Also fixed: assistant content must be array-of-blocks not plain string (silent
JS crash otherwise).
- LiteLLM: add pass_through_endpoints for /v1/responses → claude-max-bridge
- Storage, volumes, architecture docs reconciled (Vaultwarden → local zfs,
Pocket-ID backup, WAL-G fix, apps/ decommission, Nextcloud CIFS→NFS)
- Add ideas/, proxmox-memory-audit.md, llm-benchmark.md (new docs this session)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
App-password generated via occ for fkrebs@nucli.de. Gateway patched
to persist enabled flag from config.json across restarts.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Replaces outdated setup-only guide with current state:
- Full 22-route table with upstream IPs (audited 2026-05-21)
- All routes confirmed ACME+WS+Hop enabled
- JSON template for adding new routes
- Auth posture section with open items (Tinyauth pending CT 109)
- Backup/restore instructions
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
fkrebs