docs: add Scheduled Tasks feature documentation

Adds a new feature page covering the Scheduled Tasks side-panel, which
lets users run any Agent or Assistant on a cron schedule, fixed
interval, or one-shot future date:

- Registers `CalendarClock` in the sidebar icon map (`lib/icons.tsx`)
  so the page's `icon: CalendarClock` frontmatter resolves correctly,
  matching the in-app tab icon.
- Registers the page in `content/docs/features/meta.json` under the
  *Agentic AI* section.
- Documents Redis as the only required infrastructure (BullMQ-backed)
  and links to the dedicated Redis configuration page.
- Covers target/prompt configuration, the three trigger types (cron,
  interval, date) with the Quick presets dropdown and inline
  friendly-cron description, and IANA-timezone semantics including DST
  and the past-date no-op for one-shot triggers.
- Documents per-task capabilities (Web Search, File Search, Code
  Execution, MCP server picks) and notes that global `librechat.yaml`
  toggles still apply.
- Covers output handling: scheduled runs land in fresh conversations
  tagged `isScheduled: true`, hidden from the main chat sidebar, and
  surfaced via the per-task Task Runs modal; documents the Temporary
  Chat mode for non-persisted runs.
- Documents the inline task-management controls (pause/resume,
  edit-in-place, history, delete) and the post-deletion visibility
  behavior (past runs stay filtered out of the main sidebar).
- Adds a Security and Isolation section covering user-scoped data
  access and per-user execution context.
- Adds a companion local QA checklist (`QA_TESTING_GUIDE.md`) for
  verifying the Next.js/Fumadocs site builds cleanly with the new
  page.

Co-authored-by: Cursor <cursoragent@cursor.com>

Author: Golgix Auto Deploy

QA_TESTING_GUIDE.md

@@ -0,0 +1,63 @@
+# Local QA Testing Guide: LibreChat Docs (librechat.ai)
+
+This guide provides step-by-step instructions to locally test the documentation updates for the **Scheduled Tasks** feature.
+
+## Prerequisites
+
+1. Ensure you have [Bun](https://bun.sh/) installed (version 1.0+).
+2. Clone and navigate to your `librechat.ai` fork directory.
+3. Install dependencies:
+   ```bash
+   bun install
+   ```
+
+## Test Cases
+
+### 1. Build Verification
+**Goal:** Verify that the Next.js/Fumadocs site builds without any syntax or MDX errors.
+
+- Run the build command:
+  ```bash
+  bun run build
+  ```
+- **Expected Result:** The build should complete successfully without any errors about missing links, broken MDX syntax, or invalid JSON.
+
+### 2. Local Development Server & Navigation
+**Goal:** Verify the new documentation page renders correctly and is accessible from the sidebar.
+
+- Start the local development server:
+  ```bash
+  bun dev
+  ```
+- Open your browser to `http://localhost:3333` (or the port specified in the terminal).
+- Navigate to the **Features** section in the top/side navigation.
+- Look under the **Agentic AI** category in the left sidebar.
+- **Expected Result:** You should see **Scheduled Tasks** listed under `Agents API` and before `Artifacts`.
+
+### 3. Content Validation
+**Goal:** Ensure the content on the Scheduled Tasks page is formatted correctly and readable.
+
+- Click on **Scheduled Tasks** in the sidebar.
+- Verify the following sections are present and correctly formatted:
+  - **Prerequisites** (includes the Redis `.env` snippet).
+  - **Creating a Scheduled Task** (Target, Prompt, Trigger Types).
+  - **Timezone Support** (Explains IANA identifiers and UTC defaults).
+  - **Per-Task Capabilities** (Tools and MCP Servers).
+  - **Output Handling & History** (Hidden from main chat, Task Runs Modal, Temporary Chat Mode).
+  - **Security & Isolation** (User-scoped execution).
+- Check that all Markdown elements (headers, bold text, code blocks, lists) render nicely according to the site's theme.
+
+### 4. Link & Formatting Checks
+**Goal:** Check that the internal links work and formatting matches the rest of the documentation.
+
+- Verify the internal link to `[Model Context Protocol (MCP)](./mcp)` works and navigates to the MCP documentation page.
+- Run the Prettier/Lint checks (if configured) to ensure formatting compliance:
+  ```bash
+  bun run lint
+  bun run prettier
+  ```
+- **Expected Result:** No linting or formatting errors related to `scheduled_tasks.mdx` or `meta.json`.
+
+---
+
+Once all tests pass, the documentation changes are ready to be marked for review on GitHub!

content/docs/features/meta.json

@@ -8,6 +8,7 @@
     "skills",
     "subagents",
     "agents_api",
+    "scheduled_tasks",
     "artifacts",
     "code_interpreter",
     "---Search & Knowledge---",

content/docs/features/scheduled_tasks.mdx

@@ -0,0 +1,75 @@
+---
+title: Scheduled Tasks
+icon: CalendarClock
+description: Run any Agent or Assistant on a cron schedule, fixed interval, or one-shot future date — completely decoupled from the active chat session.
+---
+
+Scheduled Tasks turn an Agent or Assistant into a background worker. Define a schedule, hand it a prompt, and the task runs in its own conversation context every time it fires — with the same tools, MCP servers, web search, and code execution your interactive chats have access to.
+
+Useful for repeatable work such as:
+
+- Daily summaries (open PRs, tickets, news) delivered every morning
+- Periodic data checks via an MCP-exposed system
+- One-shot reminders at a specific wall-clock time
+- Fire-and-forget automations that post a notification via an MCP webhook tool
+
+## Prerequisites
+
+Scheduled Tasks are backed by [BullMQ](https://docs.bullmq.io/) and require a running Redis instance. The same Redis you already use for caching and resumable streams works — no extra service to operate.
+
+```bash filename=".env"
+USE_REDIS=true
+REDIS_URI=redis://localhost:6379
+```
+
+If Redis is not configured the backend boots normally but the UI's task definitions will never fire. See [Redis Configuration](/docs/configuration/redis) for hosted / cluster setups.
+
+## Creating a Task
+
+Open the **Scheduled Tasks** tab in the unified left sidebar (calendar-clock icon) and click the **+** button.
+
+### Target and Prompt
+
+- **Target Type** — `Agent` or `Assistant`.
+- **Target ID** — the unique ID of an Agent / Assistant you own.
+- **Prompt** — the message sent to the target every time the task runs. Treated identically to a chat input.
+
+### Trigger
+
+Pick one of three trigger types:
+
+- **Cron** — a standard 5-field expression, e.g. `0 9 * * 1-5` for _9 AM every weekday_. The task builder also ships a **Quick presets…** dropdown beneath the cron input (every minute, every 15 minutes, hourly, weekdays at 09:00, etc.); a friendly description is shown below the field when LibreChat recognizes the pattern. Power users can still type any 5-field expression directly — [crontab.guru](https://crontab.guru/) is a good visual helper.
+- **Interval** — a fixed delay in **milliseconds**, e.g. `3600000` for hourly.
+- **Date** — a one-shot execution at a specific datetime. Date triggers in the past simply produce no queued job.
+
+Cron and Date triggers honor an IANA timezone (e.g. `America/New_York`, `Asia/Kolkata`). New tasks default to your browser's detected zone, and cron schedules respect Daylight Saving Time transitions for the chosen zone. Interval triggers are timezone-agnostic.
+
+### Per-Task Capabilities
+
+The task builder mirrors the in-chat tool picker, scoped per task rather than inherited from the target Agent's default:
+
+- **Tools** — Web Search, File Search, Code Execution toggles.
+- **MCP Servers** — checkbox list of servers configured for your account; only the ones you tick are exposed to this task's run.
+
+This makes it easy to ship narrow, least-privilege background jobs (e.g. _"only the GitHub MCP server, nothing else"_). Per-task switches can only enable capabilities that are also enabled globally in `librechat.yaml` — if an admin has disabled Web Search, the task's Web Search toggle is a no-op.
+
+## Output Handling
+
+Every run starts a **fresh conversation** so background activity never pollutes your interactive chat history.
+
+- Scheduled-run conversations are marked `isScheduled: true` and excluded from the default conversation list.
+- The **History** (clock) icon on each task card opens the **Task Runs** modal listing every persisted run, newest first. Clicking an entry navigates to that conversation where you can read the output, inspect tool invocations, and continue manually if you want.
+- Check **Run as Temporary Chat** in the builder to skip persisting the conversation entirely — useful for fire-and-forget tasks that only need to perform a side effect. Temporary runs do not appear in the Task Runs modal.
+
+## Managing Tasks
+
+Each task card in the panel exposes inline controls for the most common workflows:
+
+- **Pause / Resume** — pause stops a recurring task without losing its configuration or history; resume re-arms the schedule.
+- **Edit** — reopens the builder pre-populated with the task's current settings. Saving atomically replaces the underlying schedule.
+- **History** — opens the Task Runs modal described above.
+- **Delete** — removes the task and its scheduled job. Past run conversations keep `isScheduled: true` and remain filtered out of the main chat sidebar even after deletion; they're still accessible via their direct conversation URL.
+
+## Security and Isolation
+
+Scheduled Tasks are strictly bound to the user who created them. A run executes in the same security context as a manual chat from that user — file ACLs, tool permissions, and MCP server credentials all apply identically. Read / update / delete API calls are scoped to the authenticated user at the data layer, so a request for another user's task ID returns `404`.

lib/icons.tsx

@@ -75,6 +75,7 @@ import {
   Import,
   Share2,
   Clock,
+  CalendarClock,
   Zap,
   // Config & structure
   Variable,
@@ -156,6 +157,7 @@ const icons: Record<string, ReactElement> = {
   Import: <Import />,
   Share2: <Share2 />,
   Clock: <Clock />,
+  CalendarClock: <CalendarClock />,
   Zap: <Zap />,
   Variable: <Variable />,
   Code: <Code />,