dank/command-center
1
0 Fork 0
Code Issues 11 Pull Requests 1 Actions Packages Projects Releases Wiki Activity

feat(models): WAN model access through the OpenWebUI door #80

Merged
dank merged 1 commits from feat/79-wan-model-access into main 2026-10-21 16:59:51 -07:00
Conversation 0 Commits 1 Files Changed 4
+290
4 changed files with 290 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

agent/models.yml
+132
View File
@@ -306,3 +306,135 @@ providers:
name: "frank-tts (dankclaw) — cloned voices"
api: openai-speech
input: [text]
# ─── WAN door ───────────────────────────────────────────────────────────
# The only block in this file that answers OFF the home network. Every
# provider above is a LAN IP, and `modelRoles` (config.yml) still points at
# those, so away from home you must NAME one of the rows below:
# omp --model 'openwebui (franksdank.cloud)/frank-dragoon:medium'
# The door is the OpenWebUI vhost `ai.franksdank.cloud` — Caddy → OWU on
# Dankclaw:3005, Anubis-exempt because OWU enforces its own auth, public A
# record present (LAN clients resolve 192.168.0.143 via Pi-hole, measured
# 2026-10-21). It exposes an OpenAI-compatible surface at /api/v1 and
# proxies to the SAME engines as the LAN rows — no second inference stack,
# so every engine-side law below still holds through it.
# CREDENTIAL LAW: the key is not in this file. `apiKey` accepts a
# command-backed value (`!cmd`, resolved by src/config/resolve-config-value.ts:
# stdout trimmed, cached for the process lifetime; anything else is tried as
# an env-var name, then taken literally), so the secret lives in
# ~/.config/secrets/openwebui_api_key, mode 600, like every other credential
# on these boxes. A literal `sk-…` here would trip the pre-commit shape gate
# and then live in every agent context on every box that pulls this repo.
# Rotate: rewrite the secret file, then RESTART omp — the `!cmd` result is
# cached per process, so a live session keeps serving the old key.
# WHAT THE DOOR SHOWS (probed 2026-10-21 with this key): /api/v1/models
# answers frank-dragoon and frank-zealot only. frank-archon is NOT published
# in OpenWebUI's picker and the API answers 400 "Model not found" for it, so
# the fleet's biggest endpoint has no WAN path until it is published
# server-side on Dankclaw — a fleet change, not a harness change.
# CONTRACT PROBED THROUGH THE DOOR (2026-10-21): streaming SSE carries
# `reasoning_content` deltas and `tool_calls` deltas, finish_reason
# `tool_calls`, and a final `usage` under stream_options.include_usage;
# `image_url` content parts reach both vision towers (a 64 px red PNG comes
# back "Red" from each); and `chat_template_kwargs` IS forwarded — proven on
# frank-zealot, where enable_thinking false gives reasoning 0 chars and true
# gives a real thinking block on the same prompt. Every compat block below is
# therefore a copy of its LAN twin, never a simplification: dropping the
# kwargs here would quietly change thinking behaviour on the WAN path.
# Chat completions made with an API key create NO OpenWebUI chat row (the
# chat list is unchanged after probing), so an off-network session leaves no
# trail in the family's chat history.
# NOT BEHIND THIS DOOR: speech. OWU does route /api/v1/audio/speech and it
# does reach frank-tts:7772, but OWU substitutes its own voice enum first —
# a cloned pack name comes back 400 from the upstream. `modelRoles.speech`
# stays on `frank-tts (dankclaw)`; do not repoint it here.
"openwebui (franksdank.cloud)":
baseUrl: https://ai.franksdank.cloud/api/v1
api: openai-completions
# Quoted because an unquoted leading `!` is a YAML tag, not a command.
apiKey: "!cat ~/.config/secrets/openwebui_api_key"
models:
- id: frank-dragoon
# Same halogen flash-server as `llamacpp (dankclaw)` above, one hop
# further out; the picker row and the engine share this id, so there
# is no alias to keep in sync.
name: "frank-dragoon (WAN)"
# Off-network the compaction hub must be reachable too: the LAN
# frank-zealot row is a dead address from anywhere but the house.
# Same engine, same slot arithmetic, so the keepRecent/threshold
# math in the LAN rows transfers untouched.
compactionModel: openwebui (franksdank.cloud)/frank-zealot
# Inherited from the engine, not invented: 200,000 under the native
# 262,144 ceiling (the 400-zone law — never sit at the ceiling). The
# proxy neither widens the window nor truncates: an over-ceiling
# request is still a 400.
contextWindow: 200000
# Same clamp arithmetic as the LAN row (see frank_dragoon_unit.bats);
# HALOGEN_MAX_TOKENS_CAP on the engine is unchanged.
maxTokens: 45056
reasoning: true
input: [text, image]
compat:
# Verified through the door: reasoning arrives as
# `reasoning_content` in the deltas, not folded into content.
reasoningContentField: "reasoning_content"
# `:off` must be a real switch, and it is: this engine honours
# reasoning_effort "none"; the harness default would only send the
# lowest effort, which still thinks.
reasoningDisableMode: none-effort
# Identity map, same reason as the LAN row — flash-server lists
# minimal|low|medium|high|xhigh|max|none and 400s anything else.
reasoningEffortMap:
minimal: minimal
low: low
medium: medium
high: high
xhigh: xhigh
max: max
# Pinned so a server-side HALOGEN_TOP_K edit cannot change this
# session's recipe; the engine accepts all three keys.
extraBody:
temperature: 1.0
top_p: 0.95
top_k: 20
- id: frank-zealot
# The 3090 endpoint, same llama.cpp engine behind the proxy. Stays
# the widest slot in the file, so it is the WAN compaction target.
name: "frank-zealot (WAN)"
compactionModel: openwebui (franksdank.cloud)/frank-zealot
# PER-SLOT window: llama.cpp divides --ctx-size across slots and
# answers a longer prompt with a 400, never a truncation. Follows
# every reslice on the engine (3 × 200,704 today).
contextWindow: 200704
# Prompt + thinking budget inside the slot; compaction fires at 85 %
# of the window on this row set (see the LAN twin for the full
# worst-case arithmetic).
maxTokens: 16384
reasoning: true
input: [text, image]
compat:
reasoningContentField: "reasoning_content"
# LOAD-BEARING, exactly as on the LAN row: the Bonsai-2 template
# accepts ONLY xhigh|medium|low and raises on anything else, so an
# unmapped harness level is a hard failure on the role's first
# request. Probed 2026-10-21 that the mapped value, not the
# harness-level, is what reaches the template through this proxy.
reasoningEffortMap:
minimal: low
low: low
medium: medium
high: xhigh
max: xhigh
# The payload's own sampling recipe plus the thinking switches.
# chat_template_kwargs survives the proxy (verified), so
# preserve_thinking keeps prior assistants' thinking blocks in
# history for multi-turn tool traffic — same contract as at home.
extraBody:
temperature: 1.0
top_p: 0.95
top_k: 20
min_p: 0.0
presence_penalty: 0.0
chat_template_kwargs:
enable_thinking: true
preserve_thinking: true
omp-docs/core/index.md
+1
View File
@@ -15,5 +15,6 @@ These files define the structural conventions that govern the agent system itsel
* [dankvoice](dankvoice.md) — Voice output is the `dank/dankvoice` package (`pi install git:`); this page keeps fleet wiring and the cutover procedure.
* [OMP Status Line Customization](statusline.md) — How presets, composer shapes, and segment arrays actually interact; the fleet's tuned bar. (moved from /pi/, 2026-10-19)
* [OMP Subagent Limits & Fan-out Policy](subagent-limits.md) — Live caps: concurrency 2, no nested spawns (the `maxRecursionDepth: 0` trap), per-endpoint in-flight ceilings, isolation policy. (moved from /pi/, 2026-10-19)
* [WAN Model Access](wan-model-access.md) — Driving the fleet's models from outside the home network through the OpenWebUI door: the command, the key file and its rotation law, what the door publishes, and the off-network failure modes.
**Change log:** the bundle keeps exactly one, one level up — [Bundle Change Log](../log.md). A directory here never has its own.
omp-docs/core/wan-model-access.md
+139
View File
@@ -0,0 +1,139 @@
---
type: reference
title: WAN Model Access (ai.franksdank.cloud)
description: How to drive the fleet's models from outside the home network — the OpenWebUI door in agent/models.yml, the key file and its rotation, what the door does and does not publish, and the off-network failure modes.
tags:
- omp
- models-yml
- openwebui
- remote-access
- fleet
verified: 2026-10-21
timestamp: 2026-10-21
---
# WAN Model Access (ai.franksdank.cloud)
Every provider in `agent/models.yml` except one is a LAN address, so a laptop
that leaves the house loses the fleet. The exception is the
`openwebui (franksdank.cloud)` block: the fleet's already-public OpenWebUI vhost
speaks OpenAI-compatible HTTP at `/api/v1` and proxies to the same engines the
LAN rows point at. One hop further out, no second inference stack.
Read this page when you are off-network and a model call fails, when the key
needs rotating, or before adding another row behind the door.
## Run a session away from home
The door is added as a provider, never as a role default. `modelRoles` in
`agent/config.yml` still names LAN providers, so away from home you must name
the model explicitly:
```bash
omp --model 'openwebui (franksdank.cloud)/frank-dragoon:medium'
omp --model 'openwebui (franksdank.cloud)/frank-zealot:off'
```
The same two ids work in the `/model` picker inside a running session — the
provider label is what disambiguates them from the LAN rows, since the wire ids
are identical by design (OpenWebUI's picker row and the engine share the id, so
there is no alias to keep in sync).
## The key
`apiKey: "!cat ~/.config/secrets/openwebui_api_key"` — a **command-backed**
value, resolved by `src/config/resolve-config-value.ts` in the harness: stdout
is trimmed, an env-var name is tried next, and only then is the string taken
literally. The secret itself is a 600-mode file in `~/.config/secrets/`, the
same convention as `gitea` and `kagi_api_key`.
Laws:
- **Never inline the `sk-…`.** It is a tracked file; the pre-commit shape gate
blocks `sk-…` anywhere in the staged diff, and a key that slipped through
would live in every agent's context on every box that pulls this repo.
- **Rotation is: rewrite the secret file, then restart omp.** The `!cmd` result
is cached for the process lifetime, so a long-lived session keeps serving the
old key — a stale-session failure reads exactly like a revoked key.
- The key is an **OpenWebUI** key (User menu → Settings → Account → API keys),
not a fleet endpoint key. It carries that user's model ACL, so what the door
shows is whatever the picker shows that user.
- Every other workstation needs its own copy of the file; nothing in the repo
carries the value.
## What the door publishes
Probed 2026-10-21 with the fleet key.
|Behind the door|Notes|
|---|---|
|`frank-dragoon`|halogen flash-server, Dankclaw iGPU — same row as the LAN `llamacpp (dankclaw)`|
|`frank-zealot`|llama.cpp on Dankclaw's 3090 — same row as `llamacpp (dankclaw-3090)`; also the WAN compaction target|
|`frank-archon`|**not published.** `/api/v1/models` omits it and the API answers `400 "Model not found"`. Dankserver's endpoint has no public face at all. Giving it a WAN path is a fleet change on Dankclaw (publish it in the OpenWebUI picker), not a harness change|
|`frank-tts`|route exists, contract does not. OpenWebUI forwards `/api/v1/audio/speech` to `frank-tts:7772` but substitutes its own voice enum first, and a cloned-pack name comes back `400` from the upstream. `modelRoles.speech` stays on the LAN `frank-tts (dankclaw)` row|
## Contract verified through the proxy
Each of these was measured end-to-end on 2026-10-21 — they are why the compat
blocks in the WAN rows are copies of their LAN twins rather than simplifications:
- Streaming SSE carries `reasoning_content` **and** `tool_calls` deltas,
`finish_reason: tool_calls`, and a final `usage` under
`stream_options.include_usage`.
- `image_url` content parts reach both vision towers.
- `chat_template_kwargs` is forwarded — proven on `frank-zealot`, where
`enable_thinking: false` yields zero reasoning and `true` yields a real
thinking block on the same prompt. Drop the kwargs from a WAN row and you
silently change thinking behaviour on the WAN path.
- API-key completions create **no** OpenWebUI chat row: the chat list is
unchanged after a probe, so an off-network session leaves nothing in the
family's chat history.
Context windows, `maxTokens`, sampling and effort maps are inherited from the
engines and documented on the LAN rows and in the fleet vault
(`serverAppConfigs-docs/LLM/`). The proxy neither widens a window nor truncates:
an over-ceiling request is still a `400`.
## Off-network failure modes
- **`Connection refused` / long timeout on a LAN id.** Not a service outage —
you are not home. Switch to the WAN id.
- **`Model not found` for `frank-archon`.** Expected; see the table above.
- **Compaction wants a LAN address.** Each WAN row sets `compactionModel` to
the WAN `frank-zealot`, but `compaction.remoteEndpoint` in `agent/config.yml`
is a LAN URL. It is last in `methodOrder`, so the other methods run first and
the remote attempt just fails — leave it unless off-network compaction turns
out to need it.
- **Roles still resolve to LAN.** A full off-network session that must not
depend on hand-picked models needs `retry.fallbackChains` entries pointing
the LAN defaults at the WAN rows; that is a deliberate `config.yml` change,
not part of adding rows here.
## Verify the door
```bash
# Key + model list (LAN clients resolve 192.168.0.143 via Pi-hole; the WAN
# A record is what makes this reachable from elsewhere).
curl -s https://ai.franksdank.cloud/api/v1/models \
-H "Authorization: Bearer $(cat ~/.config/secrets/openwebui_api_key)" | jq -r '.data[].id'
# Registry loaded, WAN rows present, no validation failure (a failed load
# silently disables EVERY custom provider — see
# agent/models.yml's schema-law comment):
omp models --no-extensions | grep -E 'franksdank|validation failed'
```
A scratch-HOME run is the only real proof that a harness change reaches the
door: copy `models.yml` + `config.yml` to `$TMP/.omp/agent/`, symlink
`$TMP/.config/secrets` at the real one (a `~` in a `!cmd` value expands to
`$HOME`), rewrite the LAN baseUrls to a dead port so a success cannot be
faked, then `HOME=$TMP omp --model 'openwebui (franksdank.cloud)/…' -p …`.
Delete the scratch tree afterwards — it holds endpoint URLs.
## Pointers
- Door topology: [Caddy](/caddy/index.md), [System hosts](/system/index.md).
- Model registry law and per-provider limits: [Core config](index.md),
[OMP Subagent Limits](subagent-limits.md).
- Server-side record: `serverAppConfigs-docs/Dankclaw/OpenWebUI.md`
(that repo's vault, not this bundle).
omp-docs/log.md
+18
View File
@@ -24,6 +24,24 @@ long form.
## 2026-10-21
### The WAN door into the model registry
`openwebui (franksdank.cloud)` joins `agent/models.yml` — the already-public
OpenWebUI vhost as an OpenAI-compatible provider, so a workstation outside the
house still reaches `frank-dragoon` and `frank-zealot` while every LAN row stays
untouched. The credential is command-backed
(`apiKey: "!cat ~/.config/secrets/openwebui_api_key"`, 600-mode file): a literal
`sk-…` in a tracked file is a key in every agent context on every box that pulls.
Laws: **OpenWebUI forwards `chat_template_kwargs`**, so a proxied row copies its
LAN twin's compat block instead of simplifying it — proved on `frank-zealot`,
`enable_thinking` false giving zero reasoning and true a real block on one
prompt; **the picker IS the API's model list**, `frank-archon` answers `400
"Model not found"` because it was never published, so a WAN path for it is a
fleet change on Dankclaw and not a harness one; and **a `!cmd` value is cached
for the process lifetime**, so rotating a key without restarting omp leaves the
old key working. Page: [core/wan-model-access.md](core/wan-model-access.md).
### the audit grows an advisory table, and corrects three of its own numbers
`omp-docs/reports/fleet-security-audit-2026-10-21.html` folds in the two per-host vulnerability