The unit re-applied --https=443 every 5s and fought Mattermost for
that port. llama is served on :10000, managed outside Nix.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
OpenCode talks to https://straylight.scylla-hammerhead.ts.net/v1, which is
Tailscale Serve on :443 proxying localhost:11434. Re-apply the serve
config on a timer so cannot leave the API dark.
whisper.cpp large-v3-turbo on :11435 (CPU, OpenAI transcriptions path)
so ASR does not take GTT from llama-server. Flash-Next is the cygnal
IQ4_XS-NGQ4 GGUF (~98 GB, gfx1151), qwen4exp, mmproj pinned; unload
Laguna before loading.
Managed /etc/opencode/opencode.json points at straylight:11434 with
the router model ids, Laguna as default, Ornith 1.5 9B as small_model,
and a vision subagent on Qwen3-VL-8B abliterated.
Ornith 1.5 9B is a dense text model, not vision. Replace that role with
prithivMLmods Qwen3-VL-8B Instruct abliterated v2 (mradermacher Q4_K_M)
and pin mmproj-url so router mode actually loads the projector.
Bind 0.0.0.0 and open 11434 on tailscale0 only, so localhost and
other tailnet machines can reach the OpenAI-compatible API. LAN
stays closed. rebuild-nixos deploys committed master, not the
working tree.
Drop the b9828 source pin and take llama.cpp from a refreshed
nixpkgs-unstable (0.4.0), which supports the Laguna and Gemma 4
architectures. Run llama-server in router mode with a preset file:
abliterated Qwen3.6-35B-A3B and Gemma 4 26B-A4B for benchmarking,
Poolside Laguna S 2.1 as the primary coding agent, and Ornith kept for
comparison. At most two models stay resident; idle models unload after
six hours. Add llama-models and llama-unload helper commands.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The TTM default caps GTT at half of RAM (~62 GiB), which is too small for
the 118B-parameter model straylight will serve. Set ttm.pages_limit and
ttm.page_pool_size to 27262976 pages (104 GiB); takes effect on reboot.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Pin herdr v0.9.0 from the upstream flake (nixpkgs-unstable only has
0.7.1) and install it on straylight to host coding-agent terminals.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Update for the 26.05 bump and host changes: 8 hosts, orion is now a
server, straylight/svc added, vade on desktop-light, MCP modules,
and the current LUKS coverage.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
BIOS boot with GRUB on /dev/xvda, MCP agent via systemd,
mc-proxy and MCNS as containers via MCP agent.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The LUKS device is named "luks-5c5e94fc-..." in hardware-configuration.nix
which already has the FIDO2 options. The "crypted" reference caused a build
error. Also fix duplicate attribute definitions and unnecessary config wrapper.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
New root CA cert issued during Metacrypt vault rebuild. Same key
usage (Certificate Sign, CRL Sign), new validity period (2026-2046).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Same key, added CRL Sign to key usage extensions. Distributed to
all nodes and NixOS system trust store.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Use nixpkgs.hostPlatform module instead of deprecated system arg to lib.nixosSystem
- Rename services.logind.powerKey to services.logind.settings.Login.HandlePowerKey
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Rootless podman deeply caches the UID in storage, subuid mappings,
and systemd sessions. Changing it destroys all container state.
Reference: log/2026-04-03-uid-incident.md
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
All nodes now list 1.1.1.1 and 8.8.8.8 as fallback nameservers after
MCNS. When MCNS is down, internal names (.svc.mcp.metacircular.net)
fail but external DNS (google.com, github.com, etc.) keeps working.
Lesson from 2026-04-03 incident: without fallbacks, MCNS failure
caused total DNS blackout including external services, forcing
Tailscale to be disabled to restore any DNS resolution.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
UID 995 conflicted with sshd on orion. Pin to 850 (the 800-899 range
is unused on all nodes and well below NixOS auto-assign range).
Pin GID to 850 as well for consistency.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
UID 995 conflicted with sshd on orion. Let NixOS auto-assign the UID
for the mcp system user. Use systemd's %U specifier for XDG_RUNTIME_DIR
instead of the hardcoded UID.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The FIDO2 crypttab options are already on the correct UUID-named device
in hardware-configuration.nix; the "crypted" name only applies to
disko-provisioned hosts (rift).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The module used explicit `config = { ... }` but also had duplicate
networking.nameservers and services.resolved.domains at the top level,
causing a NixOS module evaluation error. Merged the Tailscale nameserver
into the config block and removed the duplicates.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The master runs as an MCP-managed container, deployed via
mcp deploy mcp-master --direct. The systemd unit was a temporary
bootstrap mechanism.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Runs the MCP v2 master as a systemd service on rift. Uses
ConditionPathExists so the unit is a no-op on worker nodes
(like orion) that import mcp.nix but don't have the binary.
Starts after mcp-agent.service. Security hardened like the agent
but with ProtectHome=true (master doesn't need /run/user).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Replace fragile environment.etc.crypttab.text with
boot.initrd.luks.devices for the second SSD, matching
the pattern used for the root drive.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
kyle
