Q&A: which NapCat / LLOneBot + WS + ACP combinations have you actually run?

[happysnaker]

To help this repo feel more operationally credible, I want to publish a small deployment matrix that records which combinations have actually been exercised.

If you have a real setup, please reply with any of the following:

• NapCat or LLOneBot
• forward WS or reverse WS
• local mock agent, traecli, or custom ACP-compatible runtime
• whether you placed DeepSeek, Ollama, or another self-hosted model behind that runtime
• OS / Docker / bare-metal notes
• any token / path / ws mismatch that bit you once

A minimal reply is enough:

OneBot implementation:
WS mode:
Agent runtime:
Model layer (optional):
OS / deployment:
What went wrong once:

Useful replies may be folded back into:

• issue docs: publish tested deployment matrix for NapCat / LLOneBot and forward / reverse WS modes #6 (deployment matrix)
• FAQ / getting-started docs
• the public project page

The goal is simple: make new users hit fewer invisible setup mismatches on day 1.

[happysnaker]

Maintainer baseline answer:

Right now the clearest documented known-good path is:

• OneBot implementation: NapCat
• WS mode: reverse WebSocket
• Agent runtime: repo-built mock ACP agent
• Model layer: none for the first pass
• Deployment: Docker demo stack

Why I recommend this as the baseline:

1. it validates QQ -> OneBot -> qq-ai-bot -> ACP with the fewest external moving parts
2. it avoids blaming DeepSeek / Ollama / API keys for transport problems
3. it gives a stable first matrix row that other reports can compare against

After that, the next realistic path is:

• keep the QQ / OneBot side the same
• swap the mock agent for traecli or your own ACP-compatible runtime
• put DeepSeek or another self-hosted model stack behind that runtime

If you report another combo here, the most useful extra detail is what mismatch bit you once (token, WS direction, path, container networking, session store choice, etc.).