Polish map & UI into a command-center war-room; rewire live LLM endpoint

Map (WorldMap):
- Keep the near-black OpenFreeMap `dark` basemap (verified HTTP 200; the
  suggested dark-matter / dark-outdoors / liberty-dark variants 404 on that
  host) so the choropleth nation colors pop
- Stable per-nation fills at 0.8 opacity, thin subtle borders, and a brighter
  thicker amber border for the player nation so "this is me" is always obvious
- Hover brightens a region and shows a color-chipped owner + key stats popup;
  click opens a full dossier popup and focuses the left panel on that nation
- Transient, self-cleaning flash (pulsing amber outline) over regions whose
  ownership changed this turn
- Bottom-left legend auto-generated from the region map; per-scenario initial
  camera (Europe for the WWII scenarios, wider for Cold War / Modern / Aegis)

Command-center UI:
- New header bar: scenario title, prominent turn + date, the player nation
  (name + color chip), and a live Idle / "Simulating…" status indicator
- 3-column operations-console grid: nation dossier (left) / dominant map
  (center) / command deck + event feed (right); stacks map-first on mobile
- EventFeed: one card per event with a category icon, a colored outcome badge
  (success/failure/partial/neutral), a nation-colored accent edge, and the
  turn + date; new events slide in and flash at the top
- NationPanel: value-colored animated stat bars (red → green), territory as
  color-coded region chips, and diplomacy as a list with a colored indicator
  per relation (ally/rival/war/neutral)
- ActionPanel: a "Your move" composer with one prominent Send button and a
  spinner + "turns can take a few seconds" hint while a turn resolves
- sonner error toasts: a 502 / model-down surfaces as "AI temporarily
  unavailable — try again"

Live LLM rewire (server, previously uncommitted):
- Default endpoint https://ai.kkhost.pl/v1 (vLLM, no API key); reasoning model
  qwen3.8-27b; always send a generous max_tokens (4000) so it finishes its
  chain-of-thought before emitting content; read only `content`, ignore
  `reasoning`. docker-compose + AGENTS.md updated to match.

Tests: the deterministic Playwright e2e gains assertions for the dossier stat
bars, the map legend (all nations), and the event-card type icon + outcome
badge; it stays fully deterministic under LLM_STUB=1.
This commit is contained in:
2026-10-10 22:15:05 +00:00
parent e2936e85bb
commit e327e9aea0
15 changed files with 880 additions and 192 deletions
+14 -7
View File
@@ -123,20 +123,27 @@ history/
| Var | Default | Notes |
|---|---|---|
| `LLM_API_BASE` | `http://192.168.1.70:1111/v1` | OpenAI-compatible base URL |
| `LLM_MODEL` | `qwen3.8-27b` | model id sent in the request body |
| `LLM_TIMEOUT_MS` | `180000` | local 27B model can be slow |
| `LLM_API_BASE` | `https://ai.kkhost.pl/v1` | OpenAI-compatible **vLLM** base URL (no API key required) |
| `LLM_MODEL` | `qwen3.8-27b` | vLLM **reasoning** model — CoT in `reasoning`, final answer in `content` |
| `LLM_TIMEOUT_MS` | `180000` | the 27B reasoning model can be slow |
| `LLM_STUB` | _(unset = live)_ | `1` → deterministic stub transport (no model needed) |
> **Reasoning model:** `qwen3.8-27b` emits its chain-of-thought in the response's `reasoning`
> field and the final JSON answer in `content`. `server/llm.ts` therefore always sends a generous
> `max_tokens` (`DEFAULT_MAX_TOKENS = 4000`) so the model finishes reasoning before writing
> `content` (a too-small budget returns a null `content`), and reads only `content`, ignoring
> `reasoning`. No auth header is sent — the endpoint requires no key.
| `PORT` | `4321` | single server port |
| `DATA_DIR` | `./data` | where `history.db` lives |
| `CLIENT_DIST` | `./dist` | built client to serve (auto-found) |
## Gotchas
- **Live LLM is a runtime dependency.** If `192.168.1.70:1111` is down, turns return 502 (and the
advisor degrades to a canned list). Set `LLM_STUB=1` for a fully functional offline instance.
The map's remote OpenFreeMap style may fail to load without internet — the app still works, the
choropleth just waits for `styledata`.
- **Live LLM is a runtime dependency.** The live model is `qwen3.8-27b` on
`https://ai.kkhost.pl/v1` (a vLLM **reasoning** model, no API key). If that endpoint is down,
turns return 502 (and the advisor degrades to a canned list). Set `LLM_STUB=1` for a fully
functional offline instance. The map's remote OpenFreeMap style may fail to load without internet
— the app still works, the choropleth just waits for `styledata`.
- **Dev port override.** Vite defaults to :5173 → proxies to :4321. Both are overridable via
`VITE_PORT` / `HISTORY_API_PORT` (the Playwright config reads them) so a run can dodge a port
another project already holds. The client uses relative URLs, so it follows whichever port Vite