@huynle/hub 0.10.0 → 0.11.0

package/CHANGELOG.md

@@ -1,5 +1,83 @@
 # Changelog
 
+## 0.11.0
+
+Voice recording's Developer mode and the richer `capture.v2` recording events (below), plus the fixes and team
+tools merged since 0.10.1: a launch's `end.cleanup: keep` is honoured when an ended host launch is reconciled, and
+the applied cleanup policy is stored before a launch ends, so kept, `needs_user` and lost workspaces are no longer
+deleted later; a per-team agent message hop limit with a user-only Renew Decision when a team chain stalls (and
+orphaned stall claims expire); team member names show without graph interactions; holder-scoped grant tools and
+read-only subscription, session-model and machine/node tools over MCP; the console's Launch new member form sets
+the model, end policy and timeout; the sandbox entry publishes its PID atomically, so a second start no longer
+starts a duplicate harness. CI cancels superseded test runs without interrupting an image release and can reuse a
+PR's passing result for the same tree.
+
+### Voice recording (developer)
+- **Developer mode** (Settings → Voice, per user, off by default; user request, 2026-11-10). One more field of the
+  recording settings record (`developer`, the same revision and `PUT /api/settings/recording {developer}`): on turns
+  recording on, recording off turns it off, older consoles' `{enabled}` keep it. While a developer recording runs it
+  also keeps, behind the new hub feature `dev.v1` (start answer `developer: true`):
+  - **raw microphone audio clips** (3 s before to 2 s after) around near-misses and marks, from a ring buffer of the
+    detector's input (`audio.clip`, `PUT/GET /api/recordings/:id/clips/:n`, ≤ 200 per recording);
+  - the detector's **frame traces** around speech starts, ends and near-misses (`vad.trace`);
+  - **device and network facts** on start and on change (`device.info`: audio devices, mic track processing,
+    Network Information, visibility, wake lock, battery), **Resource Timing** of the voice routes
+    (`resource.timing`) and **console errors and warnings** (`console.log`, scrubbed, rate-limited);
+  - hub side: per turn the **model, provider, agent and context token counts** and each tool call's start and end
+    (`turn.detail`; never prompt text), **request lines** of the voice routes (`request`), and the **reply audio**
+    kept from `/api/speak` as it streams (`reply.audio`, `GET /api/recordings/:id/replies/:n`).
+  The extras count against the same caps; their audio stops first (50 MiB below the user's cap, 15 MiB below the
+  recording's; `cap.reached {what: "extras"}`), segments and events go on. Exports add `clips/` and `replies/`,
+  listed in the manifest. Turning Developer mode off stops the extras at once. The indicator says
+  "● Recording · developer"; `scripts/voice-eval.ts` reports models per turn, devices, routes, clips and errors.
+- **More of what happens is recorded**, so "I talked and nothing happened" and lost words leave a trace. New console
+  events, sent only to a hub whose recording start lists the new feature `capture.v2` (an older hub refuses a whole
+  batch with an unknown kind, value or field, so toward it the console drops them before they take a `seq`):
+  - `vad.near`: what the speech detector almost took for speech: an onset abandoned before it started
+    (`short-onset`, `voiced-ratio`) or ≥ 150 ms just under the start threshold (`below-threshold`), with its peak, the
+    threshold that applied and whether the echo guard was up (the agent speaking). At most one per 500 ms of audio
+    and 20 per minute (the rest are counted in the next summary).
+  - `vad.summary`: every 10 s of listening, the detector's frames, max and p90 speech probability, loudest frame,
+    noise floor, frames over the start and echo-guard thresholds, time under the echo guard, mode and sensitivity
+    (quiet windows after a quiet summary are skipped).
+  - `stt.received` adds `junkReason` (why a transcript was dropped as noise), `held` (the text held at that moment)
+    and `speechMs`.
+  - `stt.slow`: a transcription still unanswered after 2 s (once per segment).
+  - `mark`: a bookmark. Say **"mark"**, **"mark that"** or **"bookmark"** alone (not sent to the agent; the held
+    text waits on), or press the Voice bar's new **Mark** button (shown only while recording on such a hub). The
+    Voice bar notes "Marked".
+- The hub's `stt.result` adds the provider's `httpStatus`, `responseBytes` and `headersMs` (time to the response's
+  headers; `providerMs` stays the total).
+- **`scripts/voice-eval.ts`** reads a recording export (zip or folder) and reports, across all its recordings,
+  per-turn latency by stage (median / p90 / max, by recording kind and device), dropped segments with their reason
+  and held text, early sends, STT stalls over 3 s with the provider's share, near-misses with the echo-guard share
+  and marks with ±15 s of context (`--json`, `--no-text`). Older exports work: what they lack is reported as not
+  recorded.
+
+## 0.10.1
+
+Clicking the composer's Model chip changes the session's model. No migration is needed. Machines need the new
+plugin (and an OpenCode restart) for the chip to become clickable; with older plugins it stays display only.
+
+### Console
+- **The composer's Model chip changes the session's model** (user decision, 2026-11-10: "yes, it should change the
+  session model"). Clicking (or Enter on) "Model: …" opens the same picker as the Voice model (search over the
+  session machine's catalog, the variant step, arrows, Enter and 1–9); a pick sets the **session's own model**, so its
+  next turn and every later one run on it, whether the prompt comes from the hub console, the OpenCode TUI or a
+  Voice relay. The chip then shows it with source "chosen" (the picker marks it). The Voice chip and its picker are
+  unchanged, and a Voice pick still never changes the session's model.
+  - Mechanism: OpenCode 2.x's `session.switchModel`, which stores the model on the session (the same call the TUI
+    makes when you pick a model there); the runner uses it for every turn of that session. Last writer wins: a model
+    you then pick in the TUI's own model dialog replaces it at the TUI's next prompt. OpenCode has no way to go back
+    to "no model chosen", so the picker has no Default entry: pick the default model itself.
+  - New console route `PUT /api/sessions/:id/model {model: {providerID, modelID, variant?}}` → `{model}` (the
+    session's model now). The same access as sending a prompt: the session's owner only (foreign or unknown
+    sessions 404, session/app tokens and grant-holders 403, as for every console route). 400 for a malformed body or
+    a model or variant the session's catalog lacks, 409 `unsupported` when its spoke is older. Audited as
+    `session.model`.
+  - Protocol (additive): spoke op `setModel {sessionID, model}` → `{model: SessionModel}`, announced as the new spoke
+    feature `setModel`. Older spokes (and hubs that list no features) keep the chip display only.
+
 ## 0.10.0
 
 One release for everything since 0.9.2: the production-testing fixes, the voice update (Voice front on by default,

package/README.md

@@ -312,7 +312,7 @@ The hub serves the web app, pairs devices, enrolls machines, and does push and (
 |---|---|---|
 | **`/hub` → On this computer** | Local use; nothing to install (plugin from npm) | started by the plugin with OpenCode's own runtime (`BUN_BE_BUN=1 opencode bin/hub.ts`), or `bun` when installed |
 | **Standalone binary** | Any server or laptop, no runtime needed | `hub-<os>-<arch>` from the [releases](https://github.com/huynle/hub-releases/releases) (the source repository is private; releases are mirrored to the public `huynle/hub-releases`) |
-| **Docker** | Servers | `ghcr.io/huynle/hub` (linux/amd64, linux/arm64) |
+| **Docker** | Servers | `ghcr.io/huynle/hub` (linux/amd64, linux/arm64 for `v*` releases and `latest`; per-commit `git-…` tags from `main` are linux/amd64 only) |
 | **bunx / from source** | Bun users, development | `bunx --package @huynle/hub hub-server` · `bun bin/hub.ts` |
 | **Kubernetes (cluster mode)** | Several pods on Postgres + Valkey, rolling updates | [docs/cluster.md](docs/cluster.md), manifests in [`deploy/k8s/`](deploy/k8s/) |
 
@@ -419,6 +419,7 @@ conversation mode and utterances (403 `not_exposed` for sessions not shared with
 | `POST /api/sessions/:id/rename` | `{title}` → `{title}`: renames the session in OpenCode (`session.update`), through the process that owns it, else any other process that reports it. The title is trimmed and must be 1–200 characters (400 otherwise). Works with Voice off; the new title shows in every console at once (the spoke reports it before answering) |
 | `POST /api/sessions/:id/fork {messageID}` · `GET …/revert/preview?messageID=` · `POST …/revert {messageID, files?}` · `POST …/revert/restore` · `POST …/revert/commit` | fork before one of the session's user messages (201 `{sessionID, text, attachments}`), or stage, undo and commit a revert from it (409 `busy` while it runs); console only, audited; needs a spoke announcing `fork`/`revert` (409 `unsupported` otherwise). Contract: [docs/session-view.md](docs/session-view.md#fork-and-revert-a-message) |
 | `POST /api/sessions/:id/stop` | interrupts running descendants deepest first, then the root; idempotent. Also releases the hub's turn lock for the tree: a turn request still waiting answers `{outcome: "interrupted"}` and the next turn can start (`/interrupt` does the same) |
+| `PUT /api/sessions/:id/model {model: {providerID, modelID, variant?}}` | (feature `setModel`; console only, same access as a prompt: the session's owner) sets the session's own model: its next turn and every later one run on it, whoever prompts (the console, the TUI, a Voice relay). The spoke calls OpenCode's `session.switchModel`, which stores it on the session. Answers `{model}` (the summary's `model`, source `chosen`). 400 malformed or not in the session's catalog (`GET …/models`), 409 `unsupported` from an older spoke. Audited `session.model`. The Voice model (`…/model-prefs`) is separate |
 | `GET /api/sessions/:id/summary` | context usage, the model the session runs on (`model`: chosen, else its latest reply's, else the default; with `source`), todos, MCP status (at most 100 servers), branch and working-tree change stats (each part optional). `changes` lists at most 500 files, most-changed first, with `totalFiles` and `truncated`; its `additions`/`deletions` cover every file |
 | `GET /api/sessions/:id/context-cards` · `POST …/context-cards/:provider/refresh` | the session's [context cards](docs/context-cards.md) (`{cards, providers, enabled, configured}`), and running one provider now (`{force?}` → `{outcome: "generated", card}` · `{outcome: "skipped", reason}` · `{outcome: "failed"}`; 404 unknown provider; 429 with `Retry-After`: one per card per 30 s, 10 per user per minute). Live: SSE `{type: "contextCard", sessionID, provider, primary?, generatedAt}` (never the text) |
 | `GET /api/context-cards/primary?sessions=a,b` | the primary context cards of up to 200 sessions (`{cards: {sessionID: card}}`): the session list's second line |
@@ -818,6 +819,13 @@ the requester and may only be the requester itself or a reserved role it holds.
 the grant; approving creates it as yours, and the requester is told the grant id in a reply. One open request per
 requester, 5 an hour.
 
+`hub mcp` exposes `grant_request` (the same request route and limits), `grant_list`, `grant_show` and
+`grant_revoke` (grants held by the calling session or app; apps need `messages:read` to list/show and
+`messages:send` to revoke), and
+`grant_create` (user device credential only; an agent must request and wait for your tap). A user credential on
+the holder-scoped tools has no holder; manage all grants in the console or with `hub grant` instead. Pinning a
+profile as an app also requires a `prompt` grant on the target session, not only `profiles:assign`.
+
 **Prompts.** An agent's prompt reaches the session as the hub's, never as yours: the spoke wraps it in a provenance
 envelope (`[hub] prompt from @lead · grant g_… · msg_…`, the body between boundary lines) and admits it as a hub
 message, so it never resets the loop protection's hop chain the way your own typing does. Agents may prompt root
@@ -1101,7 +1109,12 @@ required and absolute; `mode`, `branch`, `base`, `remote`, `key`, `name`, `roles
 `cleanup`), `launches` (`state`, `label`), `launch_show` and `launch_stop` (`id`); the profile tools (`profiles`,
 `profile_show`, `profile_create`, `profile_version_create`, `profile_skill_set`, `profile_command_set`,
 `profile_assign`, …; [docs/profiles.md](docs/profiles.md#sdk-and-mcp)), the workspace, artifact and team tools
-([docs/teams.md](docs/teams.md#mcp-tools); `team_staff` takes the nested launch spec), with the same arguments
+([docs/teams.md](docs/teams.md#mcp-tools); `team_staff` takes the nested launch spec), subscription tools
+(`subscribe`, `subscriptions`, `unsubscribe`; [docs/subscriptions.md](docs/subscriptions.md#surfaces)),
+the grant tools (`grant_request`, `grant_list`, `grant_show`, `grant_revoke`, `grant_create`;
+[docs/multi-user.md](docs/multi-user.md#grants-tenancy)), `session_model_set` (user device credential only, same
+catalog checks as the console), and read-only `machines` and `launch_nodes` (user device credential only; connected
+node roots, slots, runtimes and labels; [docs/mcp-operations.md](docs/mcp-operations.md)), with the same arguments
 (`to`, `text`, `addr`, `id`, `requestID`, `formID`, `why`, `n`, `wake`, `urgent`, `all`, `project`, `role`, …). The act
 tools need an app token with scope `sessions:act` and a grant, the launch tools scope `sessions:launch` and a
 launch grant; `needs_user` results include the evaluation. `ask` takes `wait` of at
@@ -1469,7 +1482,8 @@ the Docker image, npm package and binaries ship the build). One store fed by the
   recording cancels it; each session keeps its own unsent draft), Voice mode (hands-free, see below; a toggle that
   turns Voice on first when needed; one session at a time, so turning it on elsewhere turns the first one off; it
   keeps running while you look at other sessions, which show a "Voice: <session>" pill with End) with a
-  microphone picker, and the Voice model menu (the voice front's model). The right rail (a bottom sheet on phones) has
+  microphone picker, the Voice model menu (the voice front's model) and the Model chip (the session's own model; where
+  the machine's plugin serves `setModel` it opens the same menu and sets the model every later turn runs on). The right rail (a bottom sheet on phones) has
   cards that collapse from their header (remembered per device): **Environment** (machine with
   online/offline/Connecting, project folder with the full path as tooltip, branch, and where it is open:
   TUI, Herdr or background), Context, **Connectors · N connected** (MCP servers) and Changes with a diff
@@ -1902,7 +1916,9 @@ The loop is the package `packages/voice-client` (a `VoiceTransport` over the con
   `voiceTurn.fallback`, `voiceTurn.slow` as URL query parameters (one visit) or localStorage keys. The recording
   mode records each score as `turn.predict` ([docs/voice-recording.md](docs/voice-recording.md)).
 - **Commands**: an utterance that is only "stop", "cancel", "never mind", "stop that" or "hold on"
-  calls `POST …/interrupt` and stops speech. "Pause listening" turns the mode off.
+  calls `POST …/interrupt` and stops speech. "Pause listening" turns the mode off. While the recording mode records
+  (on a hub with its `capture.v2` feature), "mark", "mark that" or "bookmark" alone bookmarks the recording instead
+  of reaching the agent ([docs/voice-recording.md](docs/voice-recording.md)).
 - **Speech out**: root replies arrive one sentence at a time while they are written (`say` with `seq`);
   the phone plays them in `seq` order per text part and drops duplicates (`SayAssembler`). A `say` without
   `seq` (a subagent's result) is split into sentences. Abbreviations, decimals, file names and

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@huynle/hub",
-  "version": "0.10.0",
+  "version": "0.11.0",
   "description": "Hub: a harness-agnostic central hub your AI agent harnesses connect to (OpenCode first). Follow, steer and stop sessions from your phone or browser, with notifications, optional hands-free Voice and a self-hostable server.",
   "license": "MIT",
   "author": "Huy Le",

package/packages/claude-code/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@huynle/hub-claude-code",
-  "version": "0.10.0",
+  "version": "0.11.0",
   "private": true,
   "description": "The Claude Code hooks spoke: per-user daemon, hook bridge, plugin dir, `hub claude` wrapper and daemon-backed MCP server (internal; shipped inside @huynle/hub, not published on its own).",
   "license": "MIT",