xmanrui/dsh-im

---

---

❤️ Sponsors

---

[!IMPORTANT] Install using the full package name: @xmanrui/dsh-im

To avoid name confusion, copy the full package name below when searching the plugin marketplace, entering it in the plugin installer, or installing from the command line.

@xmanrui/dsh-im

Choose one command: use Desktop's built-in terminal or the terminal used to start the current Harness for Web.

dsh plugin --profile desktop add -w @xmanrui/dsh-im

dsh plugin --profile web add -w @xmanrui/dsh-im

Restart Desktop or dsh web, refresh the browser for Web, then open Settings → IM Bot.

Introduction

Connect IM bots to DeepSeek Harness by scanning a QR code, using an App Manifest, or entering existing bot credentials, and let the local Harness connect outward to a public AI Office. One plugin and one settings entry manage the built-in IM channels and the AI Office Connector. iMessage uses one native Messages.app identity per macOS user account; see the iMessage channel notes.

Interface

Built-in channels

Channel	Setup	Messaging and replies
Feishu	Create a bot by QR code, or bind one with App ID + App Secret	Persistent connection for incoming messages; choose a native live process, one live process card, or step-by-step messages; each bot can enable voice interaction (optional channel capability, off by default): voice messages are transcribed into the normal text pipeline, and the answer is also sent as a voice reply
WeChat	Scan a QR code to bind a WeChat bot	Tencent iLink long polling; shows a typing indicator while Harness works, then sends the final reply in 1,800-character chunks
DingTalk	Create a bot by QR code, or bind one with Client ID + Client Secret	DingTalk Stream connection; streaming replies through AI Cards
WeCom	Create an intelligent bot by QR code, or bind one with Bot ID + Secret	Official WebSocket connection; native thinking state, tool progress, and streaming replies
QQ	Create a bot with mobile QQ QR scanning, or bind one with AppID + AppSecret	WebSocket connection; private chats show typing and receive one Markdown reply, while mentioned group chats receive only the final answer
Slack	Create an app from the bundled App Manifest, then enter a Bot Token (xoxb-) and App Token (xapp-)	Socket Mode connection; direct DM replies, mention-only channel replies, and preferred native streaming API
Telegram	Enter a Bot Token generated by @BotFather	Bot API long polling; DMs work by default and groups respond to mentions or replies, while each bot can optionally enable a private-DM allowlist; with thinking traces off, private chats, groups, and Topics edit a placeholder for progress and finalize rich replies in place, avoiding Drafts that block the Android private-chat composer; unsupported Rich delivery falls back to ordinary text
Discord	Enter a Bot Token generated in the Developer Portal	Gateway v10 connection; direct DM replies; the first mention in a server text or announcement channel creates a native Thread, where follow-up messages no longer need to mention the bot; replies stream through message edits
WhatsApp	Scan a QR code with mobile WhatsApp to link a device	WhatsApp Web connection; self-chat only by default, with optional selected-contact and open-response modes; groups require a mention or a reply to a bot message, including for the linked account; read receipt and typing indicator, with tool progress and incremental answers shown by editing one message at one-second intervals; long replies split automatically and failed edits fall back to a complete text reply
iMessage	Sign in to iMessage in macOS Messages.app and grant the local permissions described in the channel notes	Native macOS Messages.app transport for text DMs; no BlueBubbles or third-party gateway; one local iMessage identity per macOS user account
Matrix (experimental)	Enter the homeserver URL plus an access token, or a user id with a password	CS API long polling; DMs are answered directly and rooms answer when the bot is @-mentioned, with thread replies, allowlisted HTML, edit-based streaming, and image/result-file delivery; includes experimental room-message encryption/decryption, subject to the limits below

WhatsApp retains the identifiers of the latest 2,000 bot messages sent to groups, so replies to those messages still work after a restart. Replies to older messages outside that history, including unrecorded messages sent before upgrading, do not wake the bot; mention the bot or reply to one of its new messages instead.

Matrix encryption is currently for non-sensitive testing only. The default optional mode attempts to start the crypto engine and skips incoming encrypted messages if it fails; required mode refuses to connect when crypto startup fails. Interactive device verification, key backup, SSSS, and attachment-content encryption are not implemented. This release has not been validated against a live homeserver/Element setup; do not treat it as a complete end-to-end confidentiality guarantee.

Other IM platforms can be added through the same channel-adapter structure. The iMessage contribution is documented separately in the iMessage channel notes; it uses the native macOS Messages.app and requires no third-party gateway.

New Feishu and QQ bots connected through the QR flow allow all users to chat and execute built-in commands in both direct and group chats by default. Allowlists and command permissions remain configurable in each bot's access settings. Existing bots keep their saved access settings.

Feishu group chats accept messages from other bots that explicitly mention the current bot by default, with no extra setting. Messages without that mention, mentions of only other members or everyone, self-sent messages, and bot DMs are still ignored, even in all-message mode. Group allowlists and command permissions still apply. The app needs the im:message.group_at_msg.include_bot:readonly tenant scope. New apps request it through the QR flow; existing or manually connected apps can use Complete permissions or /repair in a direct chat, scan the QR code, and complete any publication or approval required by Feishu. See the Feishu message-event permissions.

Built-in channels with image input support can send JPEG, PNG, and WebP images, plus GIFs sent as image files, with optional captions to Harness. By default, original images can be up to 30 MB each, with up to 20 images per message. Originals are saved in the Session workspace; model copies are resized or compressed to 5 MB each and 20 MB total. Images that still exceed the budget, or are rejected by a non-vision model, are delivered as workspace files. These four limits can be changed under General settings → Attachments → Image input for subsequent messages. Originals follow attachment retention settings. Higher model limits increase request size, latency and model cost, and remain subject to platform and Host limits. iMessage's MVP supports text DMs only. Downloading images or files from Feishu user messages requires the im:message:readonly tenant scope, shown on the confirmation page as Read direct and group messages; Feishu currently offers no narrower image-only scope for that download endpoint. Apps created through the built-in QR flow request it by default; for existing or manually connected apps, click Complete permissions on the IM Bot settings page and scan the QR code to incrementally add that scope, im:resource for uploading bot-sent images or files, application:app_slash_command:read / write for the native command panel, and the card callback.

Shared group messages and files

WhatsApp, Feishu, Telegram, Discord, Slack, and QQ can automatically save received group messages and attachments, including messages without mentions when the platform permits delivery. Recording is enabled by default and does not itself trigger a model reply. Mention a bot in the group to search records, summarize discussions, or read files on demand.

Records are shared by group, independently of bots, AI Sessions, and workspaces. Retention defaults to a rolling 14 days, configurable under DSH Settings → IM Bots → gear icon, General settings → General. Only received content is saved; full history is not backfilled. QQ archives are currently isolated by app, and Discord threads have separate archives.

See the group history and shared files guide for all six platforms' permissions and setup steps, query examples, storage, file limits, and troubleshooting. Live tests passed for text and text-file attachments on WhatsApp, Feishu, and QQ. Telegram text passed, with files still untested; Discord and Slack await live group verification.

Results after a reply timeout

Connected IM channels share deferred task tracking. After a model reply timeout, the plugin keeps checking the original task and delivers its final text to the original chat or thread. Checks resume after a plugin restart or reconnection. /stop only stops the matching turn submitted by that chat; changing the bound Session prevents old results from being delivered there. No new setting is required, and normal replies keep their existing flow.

Channel permissions and quotas still apply. Definite send failures allow up to three attempts; uncertain delivery is retained without automatic retries to avoid duplicate messages. Recovery covers final text and terminal notices, without replaying questions, approvals, or file-tool calls. See the deferred delivery notes.

Result-file and image delivery

Built-in channels with file delivery support can return any file readable by Harness as a native channel attachment. Existing files and files created by the current task can both be sent directly. The capability is available to every connected bot by default, with no switch or per-bot allowlist, while existing text, image, streaming, command, and Session behavior remains unchanged. iMessage's MVP does not support files or attachments.

After the model calls the file-return tool, the plugin hands the specified file to the channel's native API. Images prefer native image messages; if a channel does not support or definitively rejects image delivery, the plugin falls back to a file attachment, while an uncertain result never triggers a duplicate fallback. The plugin adds no rules for file origin, creation time, workspace boundary, extension, content, count, size, or lifetime; the file only needs to exist and be readable. A channel may still reject delivery according to its own permissions, quota, file capability, or account tier, and the plugin reports that provider result.

Channel	Platform requirements
WeChat	The current binding protocol and conversation must support native file messages; the WeChat API response determines the actual range.
Feishu	Feishu's file-upload API requires a non-empty file no larger than the platform's 30 MB limit. The app needs the im:resource tenant scope (Read and upload images or other files). Apps created through the built-in QR flow request it by default; existing or manually connected apps can add it incrementally through Complete permissions or /repair in a direct chat, followed by any approval Feishu requires. The Feishu developer console currently has no separate im:resource:upload scope.
DingTalk	The app needs qyapi_base, and the bot must support file messages. The current OAPI and bot capability determine the accepted formats and sizes.
WeCom	The app needs media-upload and file-message capability; the WeCom API response determines the actual range.
QQ	The bot needs file-message capability and remains subject to QQ's daily upload quota; the bot reports when the quota is exhausted.
Slack	The Bot Token needs files:read, files:write, and reactions:write; the Workspace's current policy determines the actual file-size limit. After changing scopes, re-authorize/reinstall the App and reconnect the bot.
Telegram	The bot must be allowed to send documents in the current chat; the Bot API response determines the actual range.
Discord	Enable Message Content Intent in the Developer Portal. The bot needs Send Messages, Create Public Threads, Send Messages in Threads, and Read Message History; result-file delivery also requires Attach Files. The current account and server capability determine the actual attachment allowance.
WhatsApp	The linked session must support Document Messages; the WhatsApp/Baileys response determines the actual range.
Matrix	The homeserver must allow media uploads; its media repository configuration decides the real per-file limit. When an upload is rejected, the plugin asks you to check the media size limit and upload permissions.

AI Office Connector

Read the AI Office Connector guide

Installation

Install the published stable release of @xmanrui/dsh-im from npm (recommended). Choose the command for the application you use:

DSH Desktop: run in the current Desktop application's built-in terminal:

dsh plugin --profile desktop add -w @xmanrui/dsh-im

Restart DSH Desktop after installation.

DSH Web: run in the terminal environment used to start the current Harness:

dsh plugin --profile web add -w @xmanrui/dsh-im

Restart dsh web and refresh the browser after installation.

After restarting, open Settings → IM Bot. The top-level IM Bot entry uses order: 21 to follow Agent Presets, and the Plugins page no longer retains the old entry. Upgrading preserves existing bots, credentials, workspaces, Agent Presets, and Session bindings.

Local dsh web and DSH Desktop reuse the current Host's internal services by default: the Typert Gateway invokes the Session and Workspace controllers. No Harness address or loopback HTTP connection is required. Desktop's compatibility, extended-window, and advanced modes do not require browser access or LAN access to be enabled. An explicit channel harnessBaseUrl is retained only for legacy remote HTTP/WebSocket Harness endpoints; failed internal calls never silently switch to another Host.

To try the latest code before it is published to npm, use the GitHub-source installer instead:

npx -y github:xmanrui/dsh-im install

A GitHub-source installation fetches and builds a Git dependency directly. With pnpm 10 or newer, the profile may first need an allowBuilds entry in pnpm-workspace.yaml. Most users should prefer the stable npm release.

After installation, follow the built-in instructions on each channel page to scan a QR code or enter credentials. Secrets and Tokens are sent only to the local Harness Host and stored through its protected credential provider; status responses and bot lists never return them.

If this machine must use a forward proxy to reach Feishu, set HTTPS_PROXY to a full HTTP proxy URL before starting dsh web (for example, http://proxy:8080; lowercase https_proxy is also supported, with HTTP_PROXY accepted as a fallback), then restart the Host after changing it. Feishu registration and credential verification reuse the SDK's proxy-aware HTTP client, while the message WebSocket explicitly uses that proxy; the WebSocket path does not currently read ALL_PROXY or NO_PROXY.

If this machine cannot reach the Telegram Bot API directly, use Node.js 22.21 or newer and enable Node's environment proxy support before starting dsh web:

NODE_USE_ENV_PROXY=1 \
HTTPS_PROXY=http://proxy:8080 \
HTTP_PROXY=http://proxy:8080 \
NO_PROXY=localhost,127.0.0.1 \
dsh web

Use the proxy URL required by your network and restart the Host after changing it. If Telegram Bot Token binding reports that the Bot API cannot be reached, first check the proxy URL, Node.js version, and NO_PROXY configuration.

Default behavior	Description
Bot alias	Click the pencil beside a bot name to set an alias. Saving updates the display immediately, without restarting or reconnecting. The original name is retained; choose “Restore original name” or save an empty alias to restore it. Aliases only affect names shown in the local settings page.
Bot workspace	Each bot stores its workspace independently. New bots use $DSH_HOME/im (~/.dsh/im when unset); the directory is created when a bot or conversation selects it. Starting the plugin does not create it when all bots and conversations use other workspaces. New sessions appear under Ungrouped. Change it later from the bot card. An explicit workspace takes precedence, and dshHome overrides the DSH_HOME environment variable.
Model	Every bot in each IM channel can choose a model directly below its workspace, or follow the Host default. A change applies only to later new Sessions; send /new and then an ordinary message in the current chat to use it.
Reasoning effort	Explicitly choose an effort supported by the selected model, or follow the model default. Levels, descriptions, and defaults come from DSH. Switching models restores the new model's default effort. Each bot saves its own choice, which applies only to later new Sessions.
Agent Preset	Each bot can choose an Agent Preset on its settings card. When none is chosen, new Sessions follow the Host's agent-presets.default. A channel-level config.agentPreset is only the default for later new bots on that channel. Changing the preset never modifies or clears existing Sessions; if the current chat already has a Session, send /new and then a regular message to create one with the new selection.
Context enhancement	Open settings from a bot card to enable groups and DMs independently. Both switches default to off, including for existing bots after an upgrade.
Feishu command panel	Each Feishu bot decides, under More settings → Command panel, which commands the "/" list in the chat input offers and in what order. Follow the default keeps the shipped 34-command manifest (updated with the plugin); custom is exactly the chosen commands in the chosen order. Hiding a command only removes it from the list — typing it by hand still works — and commands added outside this manifest are never removed. Once one bot is set up, Save and copy to the other bots writes the same panel to every other bot on the channel (their own panel settings are overwritten; offline bots pick it up when they start). The bot syncs the panel in the background after saving, so a change takes a moment with many commands.
Feishu topic on mention	On by default. Mentioning the bot in a group or a private chat opens a dedicated Feishu topic under that message, and both the question and the answer stay inside it. The topic is an independent conversation in the dsh session list, and later messages in the topic belong to the same conversation. A bare "@bot" (which opens the menu) opens no topic, and messages without a mention stay in the chat's own conversation. Turn it off from the bot card's "Group" settings to return to flat replies in groups and private chats alike.
Feishu task progress display	Each Feishu bot can hide progress or use Live process, Live process card, or step-by-step messages. Live process uses Feishu's native thinking-process surface for reasoning, tool calls, and results, then sends the final answer separately. It requires Feishu desktop 7.70, mobile 7.74, or newer. Feishu's native thinking process currently cannot be posted inside a topic, so turns in a topic use the Live process card instead and keep the process and the answer in the topic.
Feishu voice interaction	Configure each Feishu bot independently under More settings → Voice on its bot card (optional channel capability, off by default). When enabled, voice messages are transcoded with ffmpeg and transcribed by DashScope qwen3-asr-flash, then processed through the normal text pipeline (commands, Sessions, model calls, and file delivery behave identically); the same turn's answer is also synthesized with qwen3-tts-flash as a voice reply to the original message. If transcription fails, the bot falls back explicitly to the existing text/image/file-only notice. The DashScope key is managed in the plugin credential store by environment variable name (secretRef, default DASHSCOPE_API_KEY) and never enters bot configuration. The ASR model, TTS model, reply voice (official qwen3-tts-flash voices such as Cherry, Serena, Bella, Chelsie, Momo, Katerina), and ffmpeg path are per-bot settings. It requires a working local ffmpeg and a key that can reach DashScope.
Session channel identity	Sessions from the local Host's IM channels and AI Office are marked with their source. The Web Session list and search results display channel logos in place of prefixes such as “WeChat ·”, preserving DSH's automatic title generation and updates. Existing Sessions receive the prefix when next loaded.

Channel prefixes are appended after DSH produces a title, preserving its complete text and automatic/manual source without pinning automatic titles or making additional model calls. Regeneration, refresh, and restarts do not stack prefixes; actual manual renames retain DSH's normal pinning behavior. This feature uses the current Host's Session events; an explicit remote harnessBaseUrl requires the plugin on the destination Host.

Logos are a dsh-im browser enhancement and require no changes to DSH. The adapter preserves original text nodes, clicks, menus, and dragging; copied titles, screen readers, and other surfaces retain the textual channel name. Unrecognized DSH page structures, unsupported browsers, or image-loading failures keep the text prefix. Unloading the plugin restores the original display.

Queueing and steering

Open a bot's More settings → General, choose New messages while a task is running, and save. Each bot has its own setting, across all 13 chat channels.

• Queue (default): process new messages after the current task finishes, with a queue receipt when they need to wait.
• Steer: add ordinary plain text to the current task for the Agent to read at its next step, with a submission receipt. There is no need to type /steer each time.

When idle, messages start a task normally. Images, files, voice messages, and quoted replies keep their existing handling. Group mention rules and permissions still apply; automatic steering requires /steer permission. Both modes retain /steer <additional instruction> for manual steering.

Saving takes effect for newly received messages immediately; already queued messages stay queued. AI Office has no ordinary chat follow-up input, so this setting does not apply to it.

Proactive delivery

IM channels with proactive delivery support can send text through a stable botId + targetId pair. Bot settings support choosing a known conversation or entering a target manually, testing the current route before saving, and copying call parameters. HTTP POST, same-Host plugins, and Connection RPC share the same target configuration and delivery core.

Saved direct-message targets also offer an opt-in Two-way Session sync switch. Once enabled, user text submitted from DSH Web/CLI and the final assistant text in that DM's current Session are mirrored back to the DM, including final assistant text from successfully completed scheduled tasks; ordinary IM prompts and /steer are not duplicated. The switch follows the current Session across /session, /new, and workspace changes. The first version supports text DMs on the current Host only; groups, Topics, Threads, and explicit remote harnessBaseUrl connections are unavailable.

Synced answers use channel-specific Markdown rendering: Feishu cards, Telegram rich text, QQ/DingTalk/Slack formatted messages, and existing WeCom bot, Matrix and Discord support. WhatsApp adapts common formatting; WeCom app and Weixin preserve the full text. Formatting fallback and long-message handling reuse channel mechanisms without adding retries for uncertain delivery.

See the Proactive Delivery Guide (简体中文) for setup steps, native fields for each supported channel, complete call examples, management endpoints, error codes, and troubleshooting.

Client panel integration

DSH versions supporting global panels show IM bots in the left sidebar, using a thin monochrome outline of the catalog's linked speech bubbles to open the full management panel. Settings → IM bots and the plugin detail page's Open IM bots action remain available. Shells can also embed the panel through the optional dshImClient service. See the client integration guide for the contract, compatibility requirements and lifecycle example.

Session → conversation lookup (host service)

Other plugins may receive only a Session id while needing to know which channel, bot, chat, or topic owns it. The host service dshIm exposes a read-only lookup for that purpose:

const info = await dshIm.describeConversation(sessionId);
// null — the Session is not owned by an IM conversation
// otherwise:
// { channel, botId, botName, conversationKey, chatType,
//   chatId, threadId, chatTitle,
//   senderId, senderName, senderUsername, senderIsBot }

• Read-only: no network access, session creation, blocking reply, or thrown error; unknown Sessions return null.
• Data comes from the plugin's state files (sessions) and bot configuration; no new core fields are required.
• chatType is group, thread, or dm. Without an author record, senderId is derived from the key only where the key names the user: Matrix direct:dm:<userId>, p2p:<id> (Feishu, Weixin, WeCom app, DingTalk), QQ c2c:<openid>, and direct:<id> on Telegram and WeCom. On Slack, Discord, WhatsApp, iMessage and others direct:<id> names the DM channel, so it is reported as chatId and senderId stays null.
• The first lookup logs one line per Session; later lookups do not repeat it.
• The bridge remembers the last inbound author's id and bot flag in the bot state, without storing message text. The display name and username are stored only when context enhancement is enabled for that chat kind (or the channel already exposes a speaker name, as Matrix does); otherwise they are null.
• Links inserted from conversation keys into numeric chat/topic fields are Telegram-only; QQ and other channels must use their native route fields and must not be treated as Telegram links.

The two binding events

A binding change is announced synchronously, so a subscriber keeps its own session → identity table without asking back:

Event	Emitted when	Payload
dsh-im/session-pre-bind	a conversation is bound to a session, before session.create	channel, botId, sessionId, conversationKey, chatType, chatId, threadId, senderId, senderName
dsh-im/session-unbind	the conversation no longer points at its session: it moved to another one, or the binding was cleared	the same fields without the author; sessionId is the old session that was detached

• Both are synchronous, and the order is detach-then-bind, so a subscriber reads one straight line.
• Every chat channel emits both events: Telegram, Discord, Email and iMessage through the shared production controller; DingTalk, Feishu, Matrix, QQ, Slack, WeCom (wecom and wecom-app), Weixin and WhatsApp assemble their own controllers but load their state through the same wrapper. office is a job connector with no conversations and emits neither event.
• A channel with its own controller must not call new StateStore(path).load() directly: createAnnouncedStateStore({ announcer, StateStore, statePath }) loads once per bot, hands back the wrapped store and the stateForSync the delivery adapter needs. Prime it at start-up, or the first prompt has nothing to answer from.
• Only committed writes announce anything: a refused write (false) or a rewrite of the same session emits no unbind.
• A subscriber has to be idempotent per (conversationKey, sessionId): a rewritten binding announces the bind again, and nothing in the protocol distinguishes that repeat from a first bind.
• Clearing one binding (clearSession) or every binding (clearSessions) emits a session-unbind per conversation.
• A throwing subscriber never breaks message handling.
• A detach does not erase the session's identity: keeping the row (with a superseded_at, for instance) still lets a scheduled task that owns that session deliver into the original chat — only "who owns this conversation" has to be answered from the latest bind.

Context enhancement

Read the context enhancement guide

Access modes

Read the access modes guide

Checking and installing updates

Read the update-checking and installation guide

Bot commands

Command	Description
/help	Show the commands and usage supported by the bot.
/menu, /m	Open the Feishu, DingTalk or Enterprise WeChat interactive menu. DingTalk arranges session, workspace, preset and model dropdowns in two columns; selections apply immediately and update the same card. WeCom requires Apply after selecting; open its menu explicitly with /m or /menu. Entering a direct chat does not open it automatically. Action buttons return their result without sending the menu again. Menus include new session, stop, compact, clear context, status and help buttons.
QQ /menu, /m	Open a button and numbered menu for session selection, workspace, mode/preset, model, new session, session list, stop, compact, steering, archived-session visibility, status and help. Lists support pagination; reply with a number when buttons are unavailable. Menus are scoped to the chat and actor and expire after 15 minutes or restart. Ordinary messages exit number selection; questions, approvals and batch input retain priority.
/new	Unbind the current chat so its next ordinary message starts a new Harness Session.
/status	Check the connection between the current bot and DeepSeek Harness.
/version	Show the version of the running dsh-im plugin.
/models	List every currently configured model with a number.
/model	Show the model and reasoning effort used by the Session bound to this chat.
/model <number or provider/model-id> [reasoning effort ID]	Switch the Session model and optionally select an effort supported by the target model.
/reasoninglist, /reasonings	Equivalent aliases that list the reasoning efforts supported by the current model.
/reasoning	Show the current Session model and reasoning effort.
/reasoning <number or effort ID>	Switch the current model's reasoning effort.
/reasoning --default	Restore the current model's default reasoning effort.
/presetlist, /presets	Equivalent aliases that list the Host's currently available Agent Presets, marking the Host default and this bot's selection.
/preset	Show this bot's Agent Preset setting for new Sessions.
/preset <number or Preset ID>	Set this bot's Agent Preset; use /preset id:<ID> for a numeric ID.
/preset --default	Clear this bot's explicit selection so later new Sessions follow the Host default.
/permissionlist	Feishu and QQ: list numbered permission presets for the bound session.
/permission [number or full ID]	Feishu and QQ: show or immediately switch the current session permissions; numeric IDs use /permission id:<ID>.
/stop	Immediately stop this chat's running task while preserving work that has not started.
/steer <additional instruction>	Inject an additional instruction into this chat's running task.
/batch	Start batch input in a direct chat and collect up to 10 text messages.
/send	Submit the collected messages, in order, as one input.
/cancel	Cancel batch input and discard its collected messages.
/repair	In a Feishu direct chat, incrementally repair the card callback and permissions required for media, group bot mentions, and the native Slash Command panel.
/compact	Immediately compact older context in the Session bound to the current chat.
/clear	Clear the model context in the Session bound to the current chat while retaining its Session binding and history; report explicitly when the current Host does not support the command.
/workspace [workspace index or absolute path], /ws [...]	View and switch the current bot's Harness workspace: without an argument it lists switchable workspaces (a tappable card on Telegram — one tap switches and the current entry is re-marked in place); an index or absolute path switches directly.
/workspacelist, /workspaces, /wsl	Legacy aliases of /workspace, equivalent to its no-argument form; native command menus keep a single workspace entry.
/sessionlist [workspace number or absolute path], /sessions [...]	Equivalent aliases that list session IDs and titles in the selected workspace and mark the session this chat is already bound to; include ungrouped sessions in the default IM directory; omit the argument to use the current workspace.
/sessionlist --all, /sessions --all	Summarize sessions across workspaces: the conversation's effective workspace comes first and is marked; 10 per workspace by default, adjustable with --limit N; unreadable workspaces are skipped and counted.
/sessionlist --limit N, /sessions --limit N	List the first N sessions in the current workspace's existing order; N must be a positive integer.
/session <Session ID, index, or title keyword>	Bind the current chat to an existing Harness session; a title keyword binds automatically only when it matches uniquely in the current conversation workspace, otherwise the full Session ID is required; archived Sessions are never selected by keyword.
/history [count]	Preview recent messages from the bound Session in a direct chat; defaults to 3, capped at 5.
Interactive question	Reply with an option number, option label, or custom text; separate multiple choices with commas. In Feishu, an image or image post from the person answering ends the pending question turn and is processed in the same Session, without finishing the remaining questions or resending the image.
Remote approval	Reply with 批准 / 拒绝 / 同意 / 不同意 / yes / no.

Web and bot approvals share one decision: the first allow-once or reject wins and retires the other prompt. Supported channels are WeChat, Feishu, DingTalk, WeCom bot/app, QQ, Telegram, Slack, Discord, WhatsApp, and Matrix; iMessage, email, and AI Office are excluded. Bot-origin approvals use the original chat. Web/Host-origin approvals require exactly one bound private chat with bidirectional sync enabled; aliases of that chat count once, while multiple chats receive a Web-only notice. Binding and access checks apply again when submitting, and permanent approval is not offered. Slack delivery from Web additionally requires im:read; update the app Manifest and reinstall the app if needed. Web remains usable when the bot cannot present an approval.

Command details

Read the command details

DingTalk menus use a shared template built into the plugin; no template setup is needed for each bot. A menu stays active for 30 minutes. Send /m again after expiry or a Host restart. Maintainers can import assets/dingtalk-menu-template.json into DingTalk's card platform to maintain the template.

Other features

Group new sessions by Agent Preset

Set groupByAgentPreset: true in the plugin Host config to opt in (off by default). A per-channel value overrides the plugin default. An explicit preset creates or reuses a named presets/<preset-id> workspace under the default IM directory for new sessions. Existing chats retain their history and location; manual workspaces, conversation overrides and Feishu directory isolation take priority. See configuration and behavior.

• Image understanding: built-in channels with image input support can send JPEG, PNG, WebP, and GIF files sent as images to Harness, with an optional text description. Originals are saved in the workspace (default 30 MB per image, 20 images per message). Model copies are resized or compressed within configurable 5 MB per-image and 20 MB total budgets, with workspace-file fallback when necessary. Configure these limits in General settings → Attachments → Image input. iMessage's MVP supports text DMs only.
• Switch workspaces from a bot card: every bot card on the settings page shows its current Harness workspace. Enter an existing absolute directory path directly or open the directory picker. Switching clears only that bot's old chat mappings; it never deletes, empties, or archives old Sessions. Replies already in progress may finish, while later messages use the new workspace.
• Default directory and ungrouped sessions: bots using the default share $DSH_HOME/im, while retaining separate chat-to-Session mappings. The plugin creates no workspace group for this directory. Use /sessions to list its ungrouped sessions and /session <Session ID>, /session N, or /session <title keyword> to bind an ordinary session. New sessions in other directories use the existing workspace grouping mechanism; switching back to the default directory creates ungrouped sessions again. Existing bot directories are not migrated automatically.
• Choose a model and reasoning effort from a bot card: every bot card in each IM channel offers model and effort controls below the workspace, with DSH-style provider groups, descriptions, and checkmarks. Select an available model and one of its supported efforts, or use the model default. Leaving the model unset follows the Host default. Settings are stored per bot and used only for later new Sessions; existing Sessions and replies already in progress are unchanged.
• Choose an Agent Preset from a bot card: every bot card can select one of the Host's existing Agent Presets, or follow the Host default. The change applies only to that bot and only to later new Sessions; existing Sessions and replies already in progress are left unchanged.
• Check the connection and send a test message: when a bot is online, clicking Check connection verifies the platform connection and sends a “DeepSeek Harness connection test succeeded” message to the bot's most recently remembered direct conversation; WhatsApp uses the account's self-chat. The test neither creates a Harness Session nor invokes the model. The bot must have received at least one direct message before it has a remembered test target; otherwise the page reports that no test conversation is available yet.
• Retry a connection or remove an integration: when a bot is offline, its card action changes to Retry connection. Use Remove integration when the bot is no longer needed. Each action affects only the selected bot and leaves other bots and channels unchanged.
• Manage multiple bots independently: a channel can have multiple connected bots. Credentials, connection state, workspace, model, Agent Preset, and chat-to-Session mappings are kept separately for every bot, so card actions do not affect sibling bots.
• Streaming replies and progress: the plugin uses each platform's available capabilities to show thinking state, tool progress, and incremental answers. Platforms without a native streaming API complete replies through message edits, card updates, or a final message.

If setup, connection, or removal fails in WeChat, Feishu, DingTalk, Enterprise WeC

…