clauderipple 0.11.0 → 0.12.0

package/dist/ui/style.css

@@ -465,6 +465,23 @@ label.check input { margin-top: 2px; }
 #phone-content code { display: block; margin-top: 6px; overflow-wrap: anywhere; }
 #phone-content > .section:first-child { margin-top: 0; }
 
+/* Ripple Remote, until a phone is paired: the connection card's shape in the accent, like the
+   laptop setup card. The QR below pairs from inside the app, so the store has its own button here. */
+.conn-card.remote-promo { --tone: var(--accent-fill); --tint: var(--accent-soft); margin-bottom: 12px; padding-bottom: 20px; }
+.remote-promo:lang(ko) { word-break: keep-all; overflow-wrap: anywhere; }
+.remote-promo .conn-icon svg { stroke-width: 1.9; }
+.remote-eyebrow { margin: 0 0 3px; color: var(--accent); font-size: var(--fs-foot); font-weight: 600; }
+.remote-points { display: grid; gap: 8px; margin: 18px 0 0; padding: 0 0 0 60px; list-style: none; }
+.remote-points li { position: relative; padding-left: 22px; color: var(--label-2); font-size: var(--fs-sub); line-height: 1.45; }
+.remote-points li::before { content: ""; position: absolute; left: 2px; top: 0.42em; width: 9px; height: 5px; border-left: 2px solid var(--accent); border-bottom: 2px solid var(--accent); transform: rotate(-45deg); }
+.remote-points strong { color: var(--label); font-weight: 600; }
+.remote-get { display: flex; flex-wrap: wrap; align-items: center; gap: 10px 16px; margin: 20px 0 0 60px; }
+.btn.large.remote-store { width: auto; gap: 8px; }
+.remote-store-icon { width: 18px; height: 18px; stroke-width: 2; }
+.remote-get .hint { flex: 1 1 200px; }
+.remote-android .login-link { white-space: nowrap; }
+.remote-next { margin: 16px 0 0 60px; padding-top: 14px; border-top: 1px solid var(--separator); color: var(--label-2); font-size: var(--fs-foot); line-height: 1.45; }
+
 /* ---- Providers ----------------------------------------------------------------------- */
 
 .provider-workspace { display: grid; grid-template-columns: 264px minmax(0, 1fr); gap: 28px; align-items: start; }
@@ -989,6 +1006,10 @@ a.cell { color: inherit; text-decoration: none; }
   .mac-side { flex: 1 1 100%; padding-left: 24px; text-align: left; }
   .mac-allow, .mac-setup-note, .mac-setup .form-field, .mac-setup-actions, .mac-setup .conn-note { margin-left: 0; padding-left: 0; }
   .mac-setup .form-field, .mac-setup-actions { max-width: none; }
+  .remote-points, .remote-get, .remote-next { margin-left: 0; padding-left: 0; }
+  .remote-get { flex-direction: column; align-items: stretch; }
+  .btn.large.remote-store { width: 100%; }
+  .remote-get .hint { flex: 0 0 auto; text-align: center; }
 
   /* Requests read as a list: model and status on top, the numbers on one quiet line below. */
   .logs-table-card { overflow: visible; }

package/docs/ARCHITECTURE.md

@@ -686,15 +686,15 @@ must preserve **Data: Artificial Analysis (artificialanalysis.ai)** attribution.
         its hits arrive as Responses events (`web_search_call`, `url_citation`), a backend the router
         does not have. DeepSeek's own endpoint and the ChatGPT provider run Anthropic's server tool;
         OpenCode Go's chat endpoint, and every `openai-compatible` provider, do not.
-      - **OpenCode Go's Anthropic endpoint does, for DeepSeek only** (corrected 2026-10-07). The
-        line above once said it did not. Re-measured with the key in `x-api-key` (a bearer there
-        answers "Missing API key"), Claude Code's own side request to `/zen/go/v1/messages` came back
-        with ten real hits and `web_search_requests: 1` for deepseek-v4.1-flash, deepseek-v4-flash
-        and deepseek-v4-pro; Kimi K3, MiniMax M3 and Qwen3.8 Flash took the tool and never searched;
-        GLM and MiMo are not served there. The preset lists those three as `anthropicSearch`, and a
-        search routed to one of them (a `smallFast` slot on OpenCode Go's DeepSeek) is relayed there
-        by `anthropicServerToolBackend`, though the model's turns stay on Chat Completions. The
-        Clients screen reads `webSearchModels` from `/api/status` to stop calling them unsearchable.
+      - **OpenCode Go's DeepSeek does not search, and never did** (corrected 2026-10-21).
+        DeepSeek's search has only ever worked on DeepSeek's own API. An entry written 2026-10-07
+        (commit 53a8db3) claimed Claude Code's side request to `/zen/go/v1/messages` came back with
+        ten real hits for OpenCode Go's DeepSeek models; that claim was wrong — the user reports it
+        never searched, and on 2026-10-21 the relay answered HTTP 200 with no
+        `web_search_tool_result` hits (`router.log`: "no results returned", twice, then a refusal).
+        The preset's `anthropicSearch` list is therefore empty, so `searchProviderFor` answers
+        nothing for those models, `/api/status` drops them from `webSearchModels`, and a search
+        routed there is refused instead of answered with invented prose.
     - **A model that cannot search does not merely decline — it invents.** Measured 2026-09-21
       against OpenCode Go's chat endpoint: `plugins: [{id: "web"}]` is accepted and ignored (HTTP
       200, no `annotations`, cache reporting no lookup), and the model answers a search-shaped
@@ -1111,10 +1111,14 @@ what we do not have yet — so that adding a provider does not start with readin
   lists them together (measured 2026-09-18). `wire` and `url` sat only on the provider,
   so that one account had to be configured as three providers — a split the user saw in
   the GUI and had to repeat in every `direct` rule. A model entry now overrides `wire`
-  (`chat`, `responses` or `anthropic`), `url` and `authHeader`, and `providerFor` in
+  (`chat`, `responses`, `anthropic` or `google`), `url` and `authHeader`, and `providerFor` in
   `config.ts` folds those into an ordinary provider at the moment a request's model is
   known, so every `provider.type` branch downstream works against one unchanged
-  contract. A provider with no override for that model is returned by identity.
+  contract. A `google` entry folds into a `google`/`api-key` provider (2026-10-21): the
+  provider's key moves from its headers to `apiKey`, the model's `url` becomes the base,
+  and the provider's `sessionHeader` and every non-auth header travel with it. The key is
+  never restated in the model entry. A provider with no override for that model is returned
+  by identity.
   `authHeader` names a convention rather than restating the key: the provider's own
   credential is re-sent under the header that model's wire expects, so there is one copy
   of the key to rotate. Each endpoint follows the API it imitates — an Anthropic-wire
@@ -1221,6 +1225,17 @@ including the 2026-10-16 Antigravity measurement in §4d-ii, are identified sepa
 - Stream endpoint `POST {base}/v1beta/models/{model}:streamGenerateContent?alt=sse`,
   the key on header `x-goog-api-key`, base default `https://generativelanguage.googleapis.com`
   (https://ai.google.dev/api/generate-content#method:-models.streamgeneratecontent).
+- **A gemini model on a relay (OpenCode Go, 2026-10-21).** `opencode-go` serves
+  `gemini-4-argon-contributor` at `https://opencode.ai/inference/go/google/v1beta/models/<id>`
+  (https://opencode.ai/docs/go/, read 2026-10-21), so the preset's entry carries
+  `wire: "google"` and that base. The transport still sends the key on `x-goog-api-key`, and
+  adds two things a relay asks for: the provider's `sessionHeader` with
+  `conversationKey(json)` — stable for one conversation, different between two, the same
+  derivation the OpenAI path uses (`providers/openai/index.ts`) — and, only when a session
+  header is set, `user-agent: clauderipple/<version>` instead of an SDK/library name. Without
+  the session header OpenCode answers `400 MissingSessionID`. Plain AI Studio providers set
+  neither and send no custom User-Agent. The provider's other headers are applied as given.
+  Not live-measured as of this date.
 - `systemInstruction` takes `{role:"user"|"model", parts:[…]}`; `contents` is the
   `user`/`model` turn list; a part is `{text}`, `{inlineData:{mimeType,data}}`,
   `{functionCall:{name,args}}` or `{functionResponse:{name,response}}`. The `role` of a
@@ -1584,6 +1599,41 @@ plan unless marked otherwise.
 | Android hotspot never recognised (2026-10-20, found by reading the code for an owner with Windows laptops and Android phones in mind): "on the hotspot" was the iPhone range 172.20.10.0/28 alone, on macOS and Windows. An Android hotspot hands out 192.168.43.x (≤ Android 10) or a random /24 (Android 11+; 192.168/16, 172.16/12, 10/8 by Android 14), so after our own failover `join()` saw no hotspot address and reported `join-failed`/`hotspot-not-visible`, while on the phone the controller read Wi-Fi and never planned the way back, and a by-hand join was never recognised | The product does not assume an iPhone. After our own join of the configured SSID is accepted by the OS, a new, non-link-local address within the settle time is the join, and its subnet (address + netmask from `os.networkInterfaces()`: no Location, no elevation) is kept in `<home>/tether.json` for that hotspot name; "on the hotspot" = iPhone range OR that network (§8). Only our own join replaces it, and a subnet equal to the Wi-Fi watched before is not remembered (logged as ambiguous) so home Wi-Fi is never read as the hotspot. Hints name both phones where the step differs. Not yet run with a real Android phone. |
 | An expired Claude login on the Mac wrote `[desktop state] …로그인이 만료되었습니다…` every few seconds per Claude app session and read the keychain (a synchronous child process) for every request, while phones learned of it only by a failed send (2026-10-15) | A login problem is one host state: logged when it starts and ends, sent to phones as `HostInfo.desktopProblem` (a banner), and the keychain is reread at most every 30 s while expired or after a 401. Per-session Remote Control errors log once per change. The host never refreshes the login itself. |
 
+### Router-owned automatic updates (2026-10-21)
+
+The always-running router owns the scheduler, not the optional tray. `autoupdate.ts` starts its
+first `update --check --json` after ten minutes and checks again every twelve hours with ±30 minutes
+of jitter. Only script and global npm installs are eligible; packaged apps, source checkouts and
+unknown layouts are checked but never automatically installed. The dashboard's **Automatic updates**
+switch saves `Config.autoUpdate`; omission means on, and changes are hot-reloaded without restarting.
+
+Installation waits for zero model calls in flight across both the proxy and OpenAI ingress, and ten
+minutes since the latest request start/end (or router startup). Busy routers are checked every five
+minutes; after twenty-four hours the wait is abandoned until the next cycle. A request arriving after
+the idle decision still follows the existing graceful drain rules above. The scheduler stops when
+that drain begins. Manual dashboard updates and automatic updates share one coordinator and the
+same detached CLI process-group path in `cli-control.ts`; the installer survives the router's exit.
+
+Before launch, `<home>/auto-update.json` is atomically written with from/to versions and attempt time.
+The CLI is pinned to that target through `CLAUDERIPPLE_PACKAGE`. The successor only records completion
+when its running version exactly matches the target; a CLI exit alone is not proof of completion.
+Failures retain the failed version, timestamp and error, suppressing automatic retries of that
+version for twenty-four hours, including across router restarts. A surviving installer PID blocks a
+second installation. Unreadable state disables installs rather than losing the retry record. Every
+scheduler decision is logged at info level. GET `/api/status` exposes `autoUpdate`, including enabled,
+install kind, available version, running state, lastUpdatedFrom/To/At and lastError.
+
+The tray consumes that status, deduplicates completed/available notifications in memory and
+`<home>/update-notifications.json`, and relaunches an outdated script/npm tray after confirmed
+completion, waiting for the installer PID to exit so its tray-runtime setup can finish first.
+Disabled automatic updates and unsupported install kinds get a click-to-update notice;
+packaged apps retain the releases-page flow. Existing versions **without this scheduler cannot be
+remotely updated into it**: their owner must first update manually through an existing supported path.
+No remotely pushed command or live installation test is part of this feature's validation.
+
+Sources: `packages/router/src/autoupdate.ts`, `cli-control.ts`, `index.ts`, `admin.ts`, `config.ts`;
+`packages/app/src/update-model.ts`, `main.ts`; scheduler and tray decision tests.
+
 ## 6. Blocked paths (measured, do not retry)
 
 - The app puts `ANTHROPIC_BASE_URL` in the CLI process env directly;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "clauderipple",
-  "version": "0.11.0",
+  "version": "0.12.0",
   "description": "Use GPT (any ChatGPT plan with Codex), DeepSeek and 400+ other models in the Claude Desktop Code tab and Claude Code while staying signed in to your Claude subscription — no third-party gateway mode. Several ChatGPT accounts that take over when one runs out, in Codex too.",
   "license": "GPL-3.0-only",
   "workspaces": [