@xingjisoft/neutrino 0.4.0 → 0.5.0

package/web/README.md

@@ -19,11 +19,13 @@ A small, dependency-free single page app that is served by the Neutrino CLI at
   access;
 - switch the interface between English and 简体中文, instantly and persistently;
 - open the same plan / token card as the terminal's `ⓘ`, clear a conversation,
-  delete a saved transcript, and review other agents' read-only history;
+  delete a saved transcript, and review or migrate other agents' history;
 - drive the running conversation: pause / resume the current turn and send new
   messages (they queue while the agent is busy);
 - pick up a saved conversation in the browser and keep chatting, even when no
-  terminal is running.
+  terminal is running;
+- run the console as a detached background service, so conversations started in
+  the browser survive closing the terminal and the CLI process.
 
 Nothing needs to be built — these are plain `.html`, `.css` and `.js` files.
 
@@ -40,6 +42,10 @@ permanently in the terminal's **Settings → Web console**. LAN access can be
 switched on and off at any time in the browser's **Settings → Web access**
 without restarting anything.
 
+To keep the console alive after the terminal exits, turn on **Settings →
+Background residency** in either workspace, or use `neutrino background` — see
+[Background residency](#background-residency).
+
 Or serve only the saved history, without a terminal:
 
 ```sh
@@ -50,6 +56,44 @@ npm run web
 
 Then open <http://127.0.0.1:8765>.
 
+## Background residency
+
+Background residency is off by default. It is a property of the process, so the
+regular and Pro workspaces share it; changing it in either page (or in the
+terminal's **Settings → Background residency**) applies immediately, including
+to browser tasks already in progress.
+
+With it on, closing the terminal or exiting the CLI does not interrupt browser
+conversations. With it off, the console follows its attached terminals and lets
+current browser tasks finish before it shuts down. The terminal's own
+conversations remain local to the terminal. Terminal and standalone launches
+share one Web console; a standalone launch reports a conflict if its requested
+port, interface or password setting differs from the running console. A ready
+service can be managed without a terminal:
+
+```sh
+neutrino background start     # detach the web console
+neutrino background status    # show its pid and address
+neutrino background open      # open it in the default browser
+neutrino background stop      # stop it
+```
+
+On Windows the service adds a notification-area icon using the same orbit badge
+as the regular console's top-left logo. Click it to open the console; its menu
+opens or stops the service. `--no-tray` suppresses it, and `NEUTRINO_NO_TRAY=1`
+does the same for automation. The service records its pid, address and a random
+management token in `~/.neutrino/background.json` (mode `0600`) and appends
+startup/stop diagnostics to `~/.neutrino/background.log`. The token protects
+`/api/background/ping` and `/api/background/stop`; it is never sent to a
+browser.
+
+When a console is mounted by the terminal and the setting is off, the first
+browser to open it is offered **建议开启后台常驻，避免误触中断任务** with a
+highlighted **开启** and **取消**. Either answer consumes the invitation for
+that CLI run — a reload does not ask again. **开启** applies immediately to the
+current browser tasks. The invitation is never shown when residency is already
+on or by a bare `neutrino web` server.
+
 ## Using it from another device
 
 The server binds to loopback by default. To open it from your phone or another
@@ -72,6 +116,11 @@ in-page switch takes over from there for the running process.
 The console can read every transcript and drive an agent with shell access, so
 keep the Web password on: anyone who can reach the port can use it.
 
+On a narrow screen the console shows one pane at a time: the conversation list,
+and then the open conversation. The back arrow in the header and the browser's
+own Back button both return to the list, and Forward reopens the conversation
+that was open.
+
 Startup prints the addresses to open:
 
 ```
@@ -179,29 +228,29 @@ login/logout routes are shared and stay unprefixed.
 
 ## Models and settings
 
-Open a conversation to see the **Provider**, **Model** and **Thinking**
-selectors. Choose a configured model and press **Apply to conversation**. The
-next turn uses that selection, and it is saved with the transcript. The same
-controls also update a conversation running in the terminal. Pause an active
-turn before applying a change; a running turn returns a clear conflict message.
+Open the model card at the lower right of the composer to choose a **Provider**
+and **Model**, and adjust **Thinking intensity** with the slider. Changes to an
+idle or paused conversation apply immediately to its next turn and are saved
+with the transcript. The same controls update a conversation running in the
+terminal. Pause an active turn before changing its model.
 
-For saved history, **Continue with model** resumes the transcript with the chosen
-selection. Changing the selectors and then using **Continue here** or sending a
-message also uses the draft selection, so a transcript whose old model was
-removed can still be continued.
+For saved history or a new workspace conversation, the card keeps the selection
+as a draft. **Continue here** or sending a message starts the conversation with
+that selection, so a transcript whose old model was removed can still continue.
 
 The **Settings** (gear) button in the sidebar opens the interface language,
 plan-first turns, the default approval mode, defaults for new conversations, the
 context compression threshold, the approval reviewer model, sound alerts, the
-extreme-risk automatic approval option, LAN access, and the Web password. Save
-settings explicitly; cancelling leaves the saved values intact. Conversation
-model changes are separate from these defaults. Models come from the same
-catalog as the terminal, and API keys stay on the server. Configure missing keys
-in the terminal model screen first. Intensity uses the same Low–Ultra sampling
-settings as the terminal, and both only offer the levels the selected model
-declares in `model.json`. The regular and [Pro](#pro-workspace) workspaces keep
-their own copy of every setting here except the approval policy, the Web password
-and LAN access, which belong to the whole process.
+extreme-risk automatic approval option, background residency, LAN access, and
+the Web password. Save settings explicitly; cancelling leaves the saved values
+intact. Conversation model changes are separate from these defaults. Models come
+from the same catalog as the terminal, and API keys stay on the server. Configure
+missing keys in the terminal model screen first. Intensity uses the same
+Low–Ultra sampling settings as the terminal, and both only offer the levels the
+selected model declares in `model.json`. The regular and [Pro](#pro-workspace)
+workspaces keep their own copy of every setting here except the approval policy,
+the Web password, LAN access and background residency, which belong to the whole
+process.
 
 ## New conversations and working directories
 
@@ -232,6 +281,21 @@ regular and Pro workspaces remember their own choice. The terminal UI stays in
 English. New conversation content, of course, remains in the language the model
 writes.
 
+## Sidebar status marks
+
+Every row in the sidebar carries one mark. A **blue dot** means the conversation
+finished and has not been opened since, a **turning ring** means a turn is
+running, and an **amber dot** means a tool call is waiting for your answer. The
+amber mark only appears while the conversation really asks for approval — with
+automatic review the reviewers answer for you, so the row keeps showing the
+running ring.
+
+The blue dot asks once: opening the row answers it, and the next turn that
+finishes asks again. The console remembers those answers in a file of its own,
+`~/.neutrino/web.json` (the regular and Pro workspaces each keep their own
+record), so the terminal UI never carries data it does not read, and a reload, a
+restart of the console or a change of port does not bring every dot back.
+
 ## History vs. live sessions
 
 A saved transcript on disk is a finished record: nothing is running, so the
@@ -272,6 +336,14 @@ the complete transcript in the conversation pane with a `read-only` badge. The
 sources are the same as the terminal's history browser; OpenCode requires Node.js
 22.13+ because it reads its SQLite database through `node:sqlite`.
 
+Click **Migrate session** at the bottom right of an opened transcript to convert
+its readable user/assistant text into a new, saved Neutrino conversation. The
+copy opens immediately in the same working directory, using the current
+Neutrino model (or the current Pro preset on `/pro`). The source is unchanged.
+Foreign system prompts, tool records and non-text attachments are not imported.
+The button is disabled while copying; a failed migration leaves the source open
+with an error so you can retry.
+
 ## Plan-first turns
 
 A browser-owned turn is planned before it runs, exactly like a turn in the
@@ -328,13 +400,17 @@ to know whether a conversation is driven by the terminal or by the server.
 | GET | `/api/fs?path=…` | Folder listing for the picker (folders only, roots, parent) |
 | POST | `/api/fs/mkdir` | Create `{ "path": "…" }` and return the new path |
 | POST | `/api/conversations` | Start `{ cwd, selection?, approval_mode? }`, returns `{ key, snapshot }` |
-| GET | `/api/settings` | Model catalog (each model lists its supported intensities), languages, public settings, and the live bind address (`web_host`, `web_urls`); no credentials |
-| POST | `/api/settings` | Save validated `{ selection?, settings? }`, including `language`, `plan_first`, `approval_mode`, `web_lan` and `web_password` |
+| GET | `/api/settings` | Model catalog (each model lists its supported intensities), languages, public settings, the live bind address (`web_host`, `web_urls`) and the background state (`background`, `background_running`, `background_url`, one-time `background_prompt`); no credentials |
+| POST | `/api/settings` | Save validated `{ selection?, settings? }`, including `language`, `plan_first`, `approval_mode`, `web_lan`, `background` and `web_password` |
+| POST | `/api/background/dismiss` | Consume the one-time background-residency invitation for this console run |
+| GET | `/api/background/ping` | Management probe of a detached service; requires `Authorization: Bearer <background.json token>` |
+| POST | `/api/background/stop` | Ask a detached service to shut down; same bearer token |
 | GET | `/api/session/:id` | One saved transcript |
 | POST | `/api/session/:id/continue` | Resume a transcript with optional `{ selection }`, returns `{ key }` |
 | POST | `/api/session/:id/delete` | Delete a saved, non-running transcript |
 | GET | `/api/agents/:agent/sessions?cwd=…` | Read-only sessions from Claude Code / Codex / OpenCode |
 | GET | `/api/agents/:agent/sessions/:id` | Read-only messages of one other-agent session |
+| POST | `/api/agents/:agent/sessions/:id/migrate` | Copy a source in `cwd` into a saved native conversation and return its live key; accepts optional `selection` or Pro `pro_preset` |
 | GET | `/api/live/:key` | Snapshot of a live session |
 | GET | `/api/live/:key/stream` | SSE stream of a live session |
 | POST | `/api/live/:key/pause` | Pause the running turn |
@@ -351,7 +427,10 @@ to know whether a conversation is driven by the terminal or by the server.
 
 Conversation and settings routes require the login cookie; see
 [Authentication](#authentication). `OPTIONS` is answered unauthenticated so
-preflights are not blocked, and it returns no data.
+preflights are not blocked, and it returns no data. The `/api/background/*`
+management routes are the exception: they exist only in a detached service and
+authenticate with the bearer token from `background.json`, because the CLI and
+the tray icon have no browser cookie.
 
 ### Approval requests