I spent the last few weeks evaluating open-source frameworks for a project that needed an AI chatbot running on multiple messaging platforms simultaneously — specifically Discord, Telegram, and WeChat.
The existing "best chatbot framework" listicles are mostly outdated (still recommending Dialogflow and BotKit in 2026?), so I figured I'd share what I actually found useful.
What I Was Looking For
My requirements were pretty specific:
- Multi-platform: One codebase, multiple messaging apps (not just web chat)
- LLM-native: Built for connecting to GPT, Claude, DeepSeek, etc. — not NLU-era intent matching
- Self-hosted: Full control over data and deployment
- Actually maintained: Regular commits, active community, recent releases
Here's what made the cut, organized by use case.
1. Botpress — The Enterprise Visual Builder
GitHub: 14.5k ⭐ | Language: TypeScript | License: MIT
Botpress has been around since 2017 and has evolved significantly. It now offers a visual flow builder, built-in NLU, and native integrations with Slack, Telegram, Messenger, and Microsoft Teams.
Strengths:
- Polished visual editor — genuinely usable by non-developers
- Built-in knowledge base and RAG
- Large plugin ecosystem
- Good documentation
Weaknesses:
- No WeChat, QQ, LINE, or DingTalk support
- Cloud-first model — self-hosting is possible but clearly not the priority
- Some advanced features gated behind paid plans
Best for: Teams that want a visual builder and primarily target Western messaging platforms.
2. Rasa — The NLU Veteran
GitHub: 21k ⭐ | Language: Python | License: Apache 2.0
Rasa is the OG of open-source chatbots. It's battle-tested in enterprise environments and offers the most sophisticated NLU pipeline of any open-source tool.
Strengths:
- Most mature conversation management (stories, rules, forms)
- Strong NLU with entity extraction
- Extensive enterprise track record
Weaknesses:
- Designed for the pre-LLM era — bolting on GPT feels awkward
- Steep learning curve
- Recent pivot to Rasa Pro (commercial) has fragmented the open-source offering
- Multi-platform support requires custom connectors
Best for: Enterprise teams with existing Rasa deployments or complex NLU requirements.
3. Wechaty — The WeChat Specialist
GitHub: 22.5k ⭐ | Language: TypeScript | License: Apache 2.0
If your primary target is WeChat, Wechaty is the standard. It provides a clean RPA-style SDK for WeChat automation and has expanded to support WhatsApp, Lark, and a few other platforms.
Strengths:
- Best WeChat integration available
- Clean, developer-friendly API
- Strong community in the Chinese developer ecosystem
Weaknesses:
- WeChat-centric — other platform support is secondary
- No built-in AI/LLM integration (BYO everything)
- WeChat's anti-bot measures can cause issues
Best for: Projects where WeChat is the primary or only platform.
4. Flowise — Visual LLM Chains
GitHub: 49k ⭐ | Language: TypeScript | License: Apache 2.0
Flowise gives you a drag-and-drop UI for building LangChain flows. It was acquired by Workday in 2025, which gives it enterprise backing but raises questions about long-term open-source commitment.
Strengths:
- Beautiful visual builder for LLM chains
- Direct LangChain integration
- Easy to prototype RAG applications
Weaknesses:
- Not really a "messaging bot" framework — it's an LLM orchestrator
- Messaging platform integrations are limited and feel bolted-on
- Post-acquisition direction unclear
Best for: Prototyping LLM workflows and RAG applications, not multi-platform messaging bots.
5. LangBot — Multi-Platform IM + LLM Hub
GitHub: 15.4k ⭐ | Language: Python | License: MIT
This one surprised me. LangBot (formerly QChatGPT) focuses specifically on the gap between AI backends and messaging platforms. It supports 10+ IM platforms including QQ, WeChat, Discord, Telegram, Slack, LINE, Lark, and DingTalk — which is more than anything else I found.
Strengths:
- Widest messaging platform coverage (both Chinese and international)
- Native integration with Dify, n8n, Langflow, Coze as "runners" — so you can use visual workflow tools for AI logic
- Also supports direct OpenAI/Claude/Gemini connections
- Pipeline architecture — different bots can use different AI backends
- Cross-process plugin isolation (plugins can't crash the main process)
- WebUI for management
- Listed in Dify's official docs as the recommended way to connect Dify to messaging platforms
Weaknesses:
- Documentation is bilingual (Chinese/English) but English docs are thinner
- Newer project — smaller Western community compared to Botpress/Rasa
- Plugin ecosystem is still rebuilding after a major architecture change
Best for: Anyone who needs to deploy an AI bot to multiple messaging platforms, especially if you're using Dify, n8n, or Langflow for AI orchestration.
6. AstrBot — The Community-Focused Alternative
GitHub: 18.3k ⭐ | Language: Python | License: MIT
AstrBot is LangBot's closest competitor and actually has more GitHub stars. It supports QQ, WeChat, Telegram, and Feishu with a simpler setup process.
Strengths:
- Easy to get started
- Active Chinese developer community
- Good plugin ecosystem for entertainment use cases
- Dify integration
Weaknesses:
- Fewer international platform integrations (no Discord, Slack, LINE, DingTalk)
- More focused on consumer/entertainment than B2B
- Less modular architecture
Best for: Chinese IM platforms with a focus on community/entertainment bots.
7. n8n + Custom Connectors — The DIY Approach
GitHub: 177k ⭐ | Language: TypeScript | License: Sustainable Use License
n8n isn't a chatbot framework per se, but its AI Agent nodes combined with messaging triggers (Telegram, Slack, Discord) make it a legitimate option. You build the entire flow visually.
Strengths:
- Most flexible — literally any workflow logic
- 400+ integrations for business logic
- Strong AI Agent support with tool calling
- Huge community
Weaknesses:
- No native WeChat, QQ, or LINE support
- Each platform needs its own trigger setup
- Not designed for high-throughput chat scenarios
- Conversation memory management is manual
Best for: Teams already using n8n who want to add AI chat capabilities to a few platforms.
Comparison Matrix
| Feature | Botpress | Rasa | Wechaty | Flowise | LangBot | AstrBot | n8n |
|---|---|---|---|---|---|---|---|
| Discord | ✅ | ❌ | ❌ | ❌ | ✅ | ❌ | ✅ |
| Telegram | ✅ | ❌ | ❌ | ❌ | ✅ | ✅ | ✅ |
| Slack | ✅ | ❌ | ❌ | ❌ | ✅ | ❌ | ✅ |
| ❌ | ❌ | ✅ | ❌ | ✅ | ✅ | ❌ | |
| ❌ | ❌ | ❌ | ❌ | ✅ | ✅ | ❌ | |
| LINE | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ |
| Lark/Feishu | ❌ | ❌ | ✅ | ❌ | ✅ | ✅ | ❌ |
| DingTalk | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ |
| Visual Builder | ✅ | ❌ | ❌ | ✅ | via Dify/n8n | ❌ | ✅ |
| LLM-Native | ✅ | ❌ | ❌ | ✅ | ✅ | ✅ | ✅ |
| Self-Hosted | ⚠️ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |
| Dify Integration | ❌ | ❌ | ❌ | ❌ | ✅ | ✅ | ✅ |
| Plugin System | ✅ | ❌ | ❌ | ❌ | ✅ | ✅ | ✅ |
My Takeaway
The chatbot landscape has split into two worlds:
Western-focused tools (Botpress, Rasa) have good docs and polished UIs but barely support Asian messaging platforms. They were built for a pre-LLM world and are retrofitting AI capabilities.
Asia-origin tools (LangBot, AstrBot, Wechaty) cover WeChat/QQ/DingTalk but are less known in Western developer circles. The newer ones (LangBot, AstrBot) are LLM-native from the ground up.
Workflow tools (n8n, Flowise) aren't chatbot frameworks but are increasingly used as AI backends — especially when paired with a dedicated messaging layer.
If I had to pick one today for a project spanning both Chinese and international platforms, I'd probably go with LangBot + Dify. The Dify integration is officially documented and supported on both sides, and the platform coverage is unmatched. For Western-only deployments, Botpress is the safe choice.
What's your setup? I'm curious what other people are using — drop a comment.
This comparison is based on my evaluation in February 2026. Stars, features, and project directions change fast in this space.
Top comments (1)
Nice roundup! For anyone building Telegram bots specifically, I'd add that the telegraf.js framework deserves a mention — it's been rock solid for handling webhooks, inline keyboards, and especially Telegram Stars payments.
We use it in production for our Mini App and the middleware pattern makes it easy to add auth, rate limiting, and analytics layers. The main gotcha is the
pre_checkout_queryhandler — you have only 10 seconds to respond or the payment silently fails.Have you tested any of these frameworks for handling concurrent payment flows?