Feature/manus api integration and bug fixes

[valentt]

Pull Request: Fix Critical Bugs and Add Manus API Integration

Summary

This PR fixes 17 critical bugs that made OpenManus non-functional and adds support for Manus API integration through a custom LLM client.

Status: ✅ All components now fully operational (Docker, Backend, Frontend, CLI)

---

Changes Overview

🔧 Infrastructure & Build Fixes (3 commits)

• ✅ fix(docker): Resolve shell script line ending issues (CRLF → LF)
• ✅ fix(deps): Add missing langchain-openai dependency
• ✅ fix(backend): Add missing __init__.py for nodes package

🔄 Backend Workflow Fixes (5 commits)

• ✅ fix(workflow): Add missing Command imports in browser and planner nodes
• ✅ fix(coordinator): Add Command import and improve response handling
• ✅ fix(workflow): Simplify graph configuration for Command-based routing
• ✅ fix(prompts): Correct template regex and add TEAM_MEMBERS support
• ✅ fix(workflow): Handle both dict and Pydantic message objects

🤖 LLM Integration Fixes & Features (2 commits)

• ✅ fix(llm): Improve PlaceholderLLM and implement actual LLM client creation
• ✅ feat(llm): Implement Manus API custom client ⭐

🎨 Frontend Fixes (3 commits)

• ✅ fix(frontend): Implement ChatInput event handlers (UI was non-interactive)
• ✅ fix(frontend): Connect UI to backend API with SSE streaming
• ✅ fix(frontend): Improve ChatDisplay styling and UX

💻 CLI Client Fix (1 commit)

• ✅ fix(cli): Update client.py for current API endpoint

📚 Documentation (3 commits)

• ✅ docs: Create comprehensive .env.example template
• ✅ docs: Update README with detailed configuration guide
• ✅ docs: Add comprehensive bug fix report (BUGFIX_REPORT.md)

🧹 Maintenance (1 commit)

• ✅ chore: Update .gitignore to exclude IDE files and planning docs

---

Detailed Bug Fixes

Critical Issues Fixed

1. Container Startup Failure

   • Error: : not found in Docker container
   • Cause: Windows CRLF line endings incompatible with Unix bash
   • Fix: Line ending conversion + .gitattributes

2. Application Crash on Startup

   • Error: TypeError: Expected a Runnable, callable or dict
   • Cause: Missing Command imports in node files
   • Fix: Added imports to all node files

3. Graph Configuration Error

   • Error: add_edge() got an unexpected keyword argument 'condition'
   • Cause: Invalid LangGraph API usage
   • Fix: Simplified graph to use Command-based routing

4. Non-functional Frontend

   • Issue: Send button did nothing
   • Cause: Missing event handlers in ChatInput component
   • Fix: Complete rewrite with proper state management

5. Frontend Not Connected to Backend

   • Issue: No API communication
   • Cause: Missing fetch implementation
   • Fix: Implemented SSE streaming with proper parsing

6. LLM Always Using Placeholder

   • Issue: Real LLM never instantiated even with API keys
   • Cause: Logic error in get_llm_by_type()
   • Fix: Added conditional client creation

See BUGFIX_REPORT.md for complete analysis of all 17 bugs.

---

🌟 New Feature: Manus API Integration

Overview

Manus API uses a task-based architecture (not OpenAI-compatible), requiring a custom implementation:

# Task Creation
POST /v1/tasks → {task_id, task_url}

# Task Polling
GET /v1/tasks/{task_id} → {status, output}

# Response Extraction
Extract assistant message from output array

Implementation

Created ManusLLM class in src/llms/manus_llm.py:

• ✅ LangChain-compatible invoke() and stream() methods
• ✅ Task creation and polling logic
• ✅ Response extraction from task output
• ✅ API_KEY header authentication (not Bearer token)
• ✅ Handles both dict and LangChain message objects

API Documentation

https://open.manus.ai/docs

---

Testing

✅ CLI Client

$ python src/client.py --task "What is 2+2?"
Response: 2+2 is 4.

✅ Web Interface

• Frontend running at http://localhost:3000
• Backend API at http://localhost:8000
• SSE streaming functional
• Message display working correctly

✅ Docker Containers

$ docker-compose up --build
✅ unified container (Backend + tools server)
✅ frontend container (Next.js dev server)

✅ Manus API Integration

• Task creation tested
• Polling and response extraction verified
• Full conversation flow working

---

Breaking Changes

None - All changes are backwards compatible and additive.

• Systems without Manus API can still use OpenAI, DeepSeek, Azure, or other providers
• PlaceholderLLM still works when no API keys configured
• Existing functionality preserved

---

Files Changed

Modified: 15 files
Created: 7 files
Total Lines Changed: ~800+

Key Files:

• src/llms/manus_llm.py (NEW) - 175 lines
• BUGFIX_REPORT.md (NEW) - 413 lines
• .env.example (NEW) - 46 lines
• src/agents/nodes/__init__.py (NEW)
• README.md - Updated configuration section
• Multiple workflow and frontend fixes

---

Documentation

New Documentation Files

1. BUGFIX_REPORT.md - Comprehensive analysis:

   • All 17 bugs with severity levels
   • Root cause analysis
   • Testing results and statistics
   • Recommendations for future work

2. .env.example - Configuration template:

   • BASIC, REASONING, VISION LLM configs
   • Azure OpenAI alternatives
   • Browser configuration options
   • Detailed explanatory comments

3. Updated README.md:

   • Detailed API key setup instructions
   • Supported LLM providers
   • Corrected CLI usage examples
   • Fixed docker-compose examples

---

Commit History

All 18 commits follow conventional commit format and are organized by functionality:

Phase 1: Infrastructure (commits 1-3)
Phase 2: Backend Workflow (commits 4-8)
Phase 3: LLM Integration (commit 9)
Phase 4: Frontend (commits 10-12)
Phase 5: CLI Client (commit 13)
Phase 6: Manus API Feature (commit 14)
Phase 7: Documentation (commits 15-16)
Phase 8: Project Docs (commit 17)
Phase 9: Maintenance (commit 18)

Each commit is independently reviewable and focused on a single concern.

---

Future Enhancements

Recommendations for follow-up work:

1. Multi-turn Conversations: Implement task continuation using taskId parameter
2. True Streaming: Consider WebSocket for real-time streaming (currently word-splitting simulation)
3. Rate Limiting: Handle Manus API rate limits gracefully
4. Vision Support: Implement attachment handling for Manus vision capabilities
5. Connector Integration: Explore Manus connectors for enhanced capabilities
6. Testing Suite: Add unit tests for ManusLLM and integration tests

---

Checklist

• ✅ All bugs fixed and tested
• ✅ Manus API integration implemented and tested
• ✅ Documentation complete and comprehensive
• ✅ Conventional commit format used
• ✅ No breaking changes
• ✅ Code follows existing patterns
• ✅ .gitignore updated appropriately
• ✅ .env excluded (only .env.example committed)

---

Questions?

Feel free to ask questions or request clarification on any of the changes. All fixes are documented in detail in BUGFIX_REPORT.md.

Thank you for considering this contribution!

[Daku3011]

Review Summary

I have reviewed the changes and tested the implementation on my end. The reported issues have been resolved successfully, and the application is now functioning correctly across the backend, frontend, Docker setup, and CLI.

The Manus API integration is a valuable addition to the project. The custom LLM client is well implemented, follows the existing architecture, and maintains backward compatibility with the supported providers. The accompanying documentation, configuration examples, and bug report also make the changes much easier to understand and review.

Overall, the changes are well organized, the commit history is clean, and the improvements to stability and usability are appreciated.

Thank you for taking the time to identify, fix, and document these issues, as well as for contributing the new Manus API support.

Status: . Ready to merge.