Chat integrations
Connect Multica agents to Feishu, Lark, Slack, DingTalk, WeCom, Telegram, or WeChat and use them from the chat tools your team already uses.
Chat integrations let the team ask an agent questions in a DM, @-mention it in a group chat, or create issues from the chat window, without opening Multica.
Feishu/Lark, Slack, DingTalk, WeCom, Telegram, and WeChat are available. WeChat accepts private messages only from the account that scans the connection QR code. They share the same session, identity, and execution mechanics but install differently.
Choose a platform
| Feishu / Lark | Slack | DingTalk | WeCom | Telegram | ||
|---|---|---|---|---|---|---|
| Installation | Generate a QR code in Multica and authorize by scanning it with Feishu | Create an app in Slack, then paste two tokens into Multica | Create an internal app and Stream-mode robot, then paste its AppKey and AppSecret into Multica | Create a smart bot with the long connection enabled in the WeCom admin console, then paste its Bot ID and Secret into Multica | Create a Bot with @BotFather, then paste its token into Multica | Scan a QR code with your own WeChat account |
| DM the agent | Supported | Supported | Supported | Supported | Supported | Only the account that scanned the QR code |
| Groups or channels | Triggered by @-mentioning the Bot | Triggered by @-mentioning the Bot | Triggered by @-mentioning the Bot | Triggered by @-mentioning the Bot | Triggered by @-mentioning or replying to the Bot | Not supported |
| Create issues | /issue message command; creates the issue from your input as-is | /issue slash command; the agent writes up the description before creating | /issue message command; creates the issue from your input as-is | /issue message command; creates the issue from your input as-is | /issue message command; creates the issue from your input as-is | /issue message command; creates the issue from your input as-is |
| List commands | /help | @Multica /help | /help | /help | /help | /help |
| Start a new Chat | /new [message] | DM: /new [message]; channel/thread: @Multica /new [message] | /new [message] | /new [message] | /new [message] | /new [message] |
| Clear the current Chat context | /clear [message] | DM: /clear [message]; channel/thread: @Multica /clear [message] | /clear [message] | /clear [message] | /clear [message] | /clear [message] |
| Connection | Platform long connection | Socket Mode | Stream mode | Platform long connection | getUpdates long polling | iLink long polling |
New connections are currently open only for mainland-China Feishu; existing international Lark connections keep working and can still be managed.
Each Bot is bound to one Multica agent. To use several agents on the same chat platform, connect a separate Bot for each.
DingTalk, WeCom, and Telegram are community-maintained: they ship in every release, but carry no official support SLA. Report problems in GitHub issues.
WeCom handles text messages today. Voice, image, and file messages get a short reply explaining that, and are not passed to the agent.
Telegram passes photos, files, video, and audio to the agent as attachments, and sends files the agent produces back into the chat, when the server has object storage configured; without it, media gets the same short unsupported reply as before. Stickers and other non-file messages get that reply and are not passed to the agent.
Step-by-step guides:
- Feishu / Lark Bot integration
- Slack Bot integration
- DingTalk Bot integration
- Telegram Bot integration
- WeChat Bot integration
How a message is processed
- Multica resolves the Bot to its workspace and agent.
- In a group or channel, only messages that explicitly @-mention the Bot continue; DMs need no mention.
- Multica verifies the sender's account binding and workspace membership.
- The message joins an agent chat session, and a run is created.
- The agent's reply goes back to the original DM or thread.
Channel messages that don't @-mention the Bot never trigger the agent and never become a turn of their own. Feishu/Lark and Telegram attach the group's recent messages as read-only context in front of the turn when someone addresses the Bot; each platform's page has the details.
/issue is a command, not a chat turn: Multica creates the issue from the message, assigns it to the agent the Bot is bound to, and posts the result back to the original chat. The command itself never appears in Multica Chat. Slack's native slash command goes through a separate asynchronous issue-creation flow.
Conversation controls
Send /help in a DM (on Slack, use @Multica /help), or @-mention the Bot with /help in a group or channel, to list the available commands: /help, /new [message], /clear [message], and /issue <title> (subsequent lines become the description). This includes WeCom and WeChat (DMs only). Multica replies directly without creating a Chat or run or changing the active Chat. On Slack, @Multica /help lists the message commands. Native /issue <description> starts a separate flow where the agent creates an issue from your description.
/new creates a new Multica Chat and routes later messages from that external conversation to it. /new <message> creates the Chat and uses the message as its first turn. The previous Chat remains stored and usable in Multica.
/clear stays in the current Multica Chat but starts a new agent-visible context. The full Chat history remains available in Multica, while the agent cannot retrieve messages from before the boundary. /clear <message> uses the message as the first turn in the new context; a bare /clear applies to the next real message.
On Slack, /new and /clear are native slash commands in a DM. A native slash-command payload does not identify a channel thread, so use @Multica /new [message] or @Multica /clear [message] inside the target thread.
Session isolation
- Feishu/Lark separates sessions by chat; later messages in the same chat continue the same session.
- Slack separates DMs by channel; in a channel, each thread holds its own session.
- DingTalk separates sessions by conversation; each DM or group continues its own session.
- WeCom separates sessions by chat; each DM or group chat continues its own session.
- Telegram separates sessions by chat; forum topics are isolated by topic.
- WeChat keeps a session for the connected Bot and the account that scanned its QR code.
Follow-ups in a channel still need a fresh @-mention. Apart from the recent context described above, the agent never reads the channel history automatically.
Account binding
For WeChat, scanning the QR code binds that WeChat account to the Multica user who started the connection. Messages use that user's workspace permissions. Scan with your own account and do not share the QR code; other accounts and group messages are not accepted.
For Feishu/Lark, Slack, DingTalk, WeCom, and Telegram, the first time a member messages the Bot, they receive an account binding link. After signing in to Multica, the platform account is linked to their membership in the current workspace.
Multica only runs the agent after binding completes. Every message re-checks account binding and workspace membership; after leaving a workspace, the Bot can no longer be used to reach it.
Account binding only confirms who the sender is. Other members on the chat platform are not automatically added to the Multica workspace.
Manage connections
Workspace owners and admins can connect or disconnect Bots; for Feishu/Lark, DingTalk, and WeChat, the agent's owner can also connect and disconnect its Bots. Regular members can view connected integrations and use the agents they have access to.
After disconnecting, the Bot stops receiving new messages. Existing Multica conversations and run records are kept.
Self-hosting
A self-hosted deployment must configure a 32-byte encryption key for each platform before Multica opens its connection entry point:
MULTICA_LARK_SECRET_KEY=<base64-encoded 32-byte key>
MULTICA_SLACK_SECRET_KEY=<base64-encoded 32-byte key>
MULTICA_DINGTALK_SECRET_KEY=<base64-encoded 32-byte key>
MULTICA_WECOM_SECRET_KEY=<base64-encoded 32-byte key>
MULTICA_TELEGRAM_SECRET_KEY=<base64-encoded 32-byte key>
MULTICA_WEIXIN_SECRET_KEY=<base64-encoded 32-byte key>These keys encrypt stored Bot credentials. See Environment variables for how to generate, store, and rotate them. Multica Cloud already has this configured.
WeCom can only send replies through the WebSocket held by one process. What happens to a reply produced on another replica depends on the realtime relay mode:
- Sharded or dual relay mode (
REDIS_URLset — the default with Redis): the reply is forwarded to the replica holding the connection and delivered, so WeCom can run on several replicas. - Legacy relay mode, or no Redis: the reply is dropped. Run the WeCom-enabled backend as a single replica in this configuration.
In every mode, a reply produced while no replica holds a live connection (all mid-reconnect) is not delivered. The replica that routed it later checks whether any replica claimed the delivery and, if none did, increments multica_wecom_outbound_dropped_total{reason="no_live_connection"}, so the metric shows how often this happens. If that loss is unacceptable, a single replica remains the most conservative deployment.
Next steps
- Feishu / Lark Bot integration — connect an agent by scanning a QR code.
- Slack Bot integration — create a Slack app and paste its tokens into Multica.
- DingTalk Bot integration — create a Stream-mode robot and paste its credentials into Multica.
- Telegram Bot integration — create a Bot with @BotFather and paste its token into Multica.
- WeChat Bot integration — scan a QR code with your own WeChat account.
- Chat — the conversation and execution model behind every Bot.