Proposal: Integrate CodeArts Agent as a Multica runtime

[lithinkit]

TL;DR

We have built codearts-multica-runtime — a bridge that integrates CodeArts Agent (华为云码道代码智能体) into Multica as a standard agent runtime, following the same pattern as dsh-multica-runtime.

What is CodeArts?

CodeArts Agent is Huawei Cloud's AI-powered coding assistant kernel, available as a VS Code extension. It exposes a local REST API (SSE-based streaming) at localhost:49041, backed by DeepSeek models (e.g., deepseek-v4-pro).

What we built

Following the same protocol (v=1) and architecture as dsh-multica-runtime:

• Uses the same JSONL frame protocol (execute, cancel, probe, list-models)
• Runs in three modes: --probe / --list-models / --stdio
• Maps CodeArts Agent REST API (session create, SSE event stream, prompt_async) to Multica runtime protocol
• Zero additional dependencies — pure Node.js 18+ stdlib

Architecture

Multica Daemon (JSONL/stdin)
  → codearts-multica-runtime (protocol bridge)
    → CodeArts Agent Kernel (localhost:49041, HTTP Basic Auth)
      → DeepSeek LLM

Project structure

codearts-multica-runtime/
├── src/
│   ├── index.ts          # CLI entry (probe/list-models/stdio)
│   ├── protocol.ts       # JSONL frame parser/serializer
│   ├── client.ts         # CodeArts Agent HTTP client
│   ├── sse.ts            # SSE event stream parser
│   ├── probe.ts          # Health check handler
│   ├── models.ts         # Model listing handler
│   ├── execute.ts        # Execute/cancel handler
│   └── env.ts            # Environment variable validation
└── tests/

Multica config

runtimes:
  - name: codearts
    command: ["node", "/path/to/codearts-multica-runtime/dist/index.js"]
    probe_args: ["--probe"]
    list_models_args: ["--list-models"]
    exec_args: ["--stdio"]
    env:
      - OPENCODE_SERVER_USERNAME
      - OPENCODE_SERVER_PASSWORD

Why this matters for Multica

1. Expands runtime ecosystem: CodeArts brings Huawei Cloud's entire developer toolchain (CodeArts IDE, deploy, artifact hosting) into the Multica multi-agent world
2. More provider choice: Users can leverage DeepSeek models through the CodeArts kernel as one more runtime option
3. Proven pattern: We strictly followed the dsh-multica-runtime architecture, so integration is low-risk
4. Enterprise-grade: CodeArts is a production-ready, Huawei Cloud-hosted service with enterprise auth and security

What we are asking

We would like to discuss upstreaming this into the Multica ecosystem — whether as an official runtime, a community-maintained runtime, or simply documentation/recognition in the runtime registry.

We are also open to:

• Code review and architecture feedback
• Suggestions for supporting MCP tool forwarding and permission approval
• Discussion on multi-kernel instance support

Current status

• Design doc: ✅ complete
• Core implementation: 🔨 in progress
• Integration tests: 📋 planned

Happy to share the full repo and design doc. Looking forward to feedback!