FAQ

5-minute answers on choosing, deploying and securing an enterprise AI assistant

Product & Positioning

What is Mule Agent? What scenarios is it for?

Mule Agent is an enterprise-grade AI assistant platform positioned as a "unified multi-IM platform for enterprises" — it connects Feishu (Lark), WeCom, WeChat, DingTalk, QQ, Teams, WhatsApp and more in one admin console.

It fits companies that use multiple IM platforms at once, need knowledge-base management, need role/department-based permissions, or have on-premises deployment or data compliance requirements.

How is it different from DingTalk's built-in AI assistant?

DingTalk's AI assistant only runs inside the DingTalk ecosystem and cannot connect to Feishu, WeCom or other platforms at the same time.

Mule Agent's core difference is unified multi-IM integration — one console manages all IMs, and employees use the AI assistant in the IM they already use, without switching apps. It suits multi-IM environments or scenarios where you want unified knowledge management across different IM conversations.

What is the difference between Mule Agent and OpenClaw / WorkBuddy?

There are three main differences:

  • DingTalk support: Only Mule Agent natively supports DingTalk; neither OpenClaw nor WorkBuddy does
  • Server-side deployment: Mule Agent runs server-side 24/7; OpenClaw and WorkBuddy both depend on a local PC
  • Enterprise features: Mule Agent has multi-tenant isolation, RBAC permissions and role-filtered knowledge bases for compliance; OpenClaw and WorkBuddy are designed for a single operator with no enterprise permission system

See the full comparison for details.

Integration & Deployment

Do Feishu/DingTalk/WeCom integrations need a public IP?

No.

Mule Agent supports two integration modes:

  • Webhook callback: use the IM platform's outbound webhook; Mule Agent runs on your intranet and receives messages via the callback URL — no public IP needed
  • Long connection (WebSocket): supported by some platforms for a direct two-way channel

Quickstart: Feishu bot in 30 minutes

Do you support on-premises deployment? Where is data processed?

Three deployment options:

  • Cloud deployment: Mule Agent runs on a cloud server; data passes through the Mule Agent server
  • Hybrid deployment: AI inference runs locally (e.g. Ollama) while IM integration and orchestration stay in the cloud; sensitive data never leaves your intranet
  • Fully on-premises: everything runs on your own servers (VPS/private IDC) — IM integration, AI inference and data storage all stay on the intranet

Hybrid and fully on-premises suit finance, healthcare, government and other industries with strict compliance requirements.

How long does deployment take?

From zero to a working Feishu integration, reference times:

  • Cloud quick trial: ~30 minutes (with a server)
  • Full Feishu integration: ~1-2 hours (including app creation, callback config, knowledge-base import)
  • Multi-IM + knowledge base: 1-3 days

See the week-1 checklist in the 30-day rollout SOP.

Data & Security

Is enterprise data secure? Can it leak?

Mule Agent's security design:

  • On-premises deployment: data stays entirely on your own servers, never through a third party
  • Tool sandbox: external tools invoked by the AI (file reads, script execution) run in an isolated environment and cannot directly reach your intranet
  • Multi-tenant isolation: data of different companies/departments is fully isolated
  • RBAC: knowledge bases can be scoped by department/role
  • Inbound filtering: incoming messages are validated to prevent prompt injection

Compliance requirements in finance and healthcare can be met with fully on-premises deployment. See the AI compliance guide for finance.

Which LLMs are supported? Can I use my own models?

Mule Agent supports many models:

  • Cloud APIs: OpenAI GPT series, Claude, Google Gemini, DeepSeek, Alibaba Qwen and more
  • Local models: open-source models via Ollama (Llama, Qwen, Mistral, etc.)
  • MCP protocol: connect any MCP-compatible tool or service

Local models suit scenarios where data must not leave the network and cost less; cloud APIs perform better and suit latency-sensitive scenarios.

How is the knowledge base behind AI answers managed?

Mule Agent has a built-in knowledge base manager:

  • Multiple formats: PDF, Word, Excel, PPT, Markdown, plain text and more
  • Vector search: documents are chunked and embedded; retrieval uses semantic matching, not keywords
  • Hybrid retrieval: vector similarity + BM25 keyword recall for higher accuracy
  • Permission linkage: knowledge bases can be filtered by department/role — different employees see different knowledge
  • Incremental updates: changed documents are re-chunked and re-embedded automatically, no full rebuild

Features & Pricing

What built-in features are there? Do I need to configure anything?

Out-of-the-box capabilities include:

  • 60+ built-in skills: calendar, email, spreadsheets, documents, code execution and more
  • 22+ IM platform adapters: covering mainstream IMs
  • Scheduled tasks (Cron): runs without a PC, fully automated
  • MCP integration: connect more tools via the MCP protocol
  • Skills orchestration: compose multi-step workflows in natural language

No coding required — everything is configured from the admin console.

Is the product multilingual? English / Traditional Chinese?

Yes.

Mule Agent supports Simplified Chinese, Traditional Chinese and English, switchable in the console. The AI's answers follow each employee's language setting.

Enterprise Selection & Management

How do I choose between Mule Agent and open-source options like OpenClaw / Hermes?

Focus on three points:

  • DingTalk support: OpenClaw doesn't support DingTalk; Mule Agent supports it natively, plus full coverage of Feishu, WeCom, WeChat and QQ
  • Enterprise-grade permissions: OpenClaw is designed for a single operator — no multi-tenancy, no RBAC; Mule Agent has multi-tenant isolation and a department/role/level permission system
  • Server-side operation: open-source options depend on a local PC and go offline with it; Mule Agent runs 24/7 on the server, and scheduled tasks don't depend on a PC

See the detailed comparison for item-by-item comparison.

Can multiple departments/subsidiaries share one system? Is data isolated?

Yes. Mule Agent supports multi-tenant isolation: data of different companies/departments is fully isolated and invisible to each other — ideal for groups with multiple subsidiaries, or service providers serving multiple clients at once.

The permission system assigns access by department/role/level, and knowledge bases can be filtered by role — different employees see different knowledge.

Can employee AI usage be tracked? How do we control costs?

Yes. The admin console provides per-user token usage statistics, showing how many tokens each employee has used — handy for cost accounting and quota management.

LLM API fees are pay-as-you-go (mainstream models ~0.1-2 RMB per 1k tokens); combined with usage stats, you can control costs down to the individual.

Is there an audit log? Can I trace who used what?

Yes. Mule Agent provides credential access audit logs: who used which credential at what time and what they did — fully recorded and traceable.

Meets internal audit and compliance requirements; issues can be pinned down to a person and a timestamp.

We're already on OpenClaw / Hermes. Can we migrate?

Yes. Mule Agent supports one-click migration: agent-driven, automatically importing Skills, Memory and configuration from OpenClaw / Hermes — data fully preserved, zero retraining for the team.

Your existing IM platforms keep working after migration; no re-training needed.

Do scheduled tasks require the PC to stay on?

No. Mule Agent runs server-side (VPS/private server), 24/7 online. Scheduled tasks (cron) execute on the server and don't depend on a personal computer — turning it off doesn't matter.

Different departments/roles can configure different scheduled tasks, each with its own context.

Can Feishu docs/sheets/Wiki be deeply integrated, not just messaging?

Yes. Mule Agent has full Feishu integration: all 5 components — Docs, Sheets, Wiki, Bitable and cloud documents — natively supported, readable and writable.

The AI can look up documents, summarize spreadsheets, organize Wiki pages and read/write Bitables — not just send and receive messages.

Troubleshooting & Support

AI answers are inaccurate/wrong. What should I do?

Common causes and fixes:

  • The knowledge base doesn't cover the question → add relevant documents and re-import
  • Documents are outdated → remove old docs, import the latest version, update vectors incrementally
  • Wrong retrieval recall → adjust chunk size, add synonyms, enable hybrid retrieval
  • Prompt guardrails are weak → add "say you don't know" and a confidence threshold

See 8 common RAG knowledge-base pitfalls.

Who supports me when something goes wrong?

Mule Agent offers commercial support:

  • Online support via Feishu/WeChat groups
  • Documentation center and usage guides
  • Enterprise SLA available by contract

Email: 278946228@qq.com

Still have questions?

Tell us your IM platforms, team size and industry — get a recommendation in as little as 30 minutes

Contact us →