Paul Cuciureanu

The docs are unclear on this: https://gitlab.com/deposition.cloud/infra/devops/devcontainer/bloat/-/blob/main/Dockerfile?ref_type=heads confusing behavior with `SANDBOX_RUNTIME_CONTAINER_IMAGE` and `SANDBOX_BASE_CONTAINER_IMAGE` after some struggle, I reverse engineered the _runtime_ container image my suggestion would be to drop support...

> I'm working on an alternative approach to perform routing by interleaving calls to different LLMs in a trajectory using a fine-tuned LLM as the router. @ryanhoangt - how about...

not sure, but if required, maybe https://github.com/sparfenyuk/mcp-proxy and https://github.com/punkpeye/mcp-proxy could come in handy

while not the same architecture as A2A, the way LangGraph goes about it conceptually might be illustrative? [Multi-agent swarms with LangGraph](https://www.youtube.com/watch?v=JeyDrn1dSUQ) imagine (in the docker enabled multi-user anything llm context)...

The current backwards-compatible friendly behaviour should be _thoroughly_ documented. I put the volumes in reverse order, assuming standard Docker volume spec behaviour, and was wondering why /workspace was pointing to...

Setting `DEBUG` and `DEBUG_RUNTIME` to `true` seems to work

this might help: https://github.com/NickSherrow/openoperator cc: @NickSherrow

I'm still not 100% sure I'm doing this "the right LangGraph way", given the spotty documentation on the topic Happy to tweak this PR, if we could simplify it further!...

Uh oh, this will be painful... good luck! I would suggest narrowing down on a multi polyglot repos workflow first so OpenHands can still gather all the context it needs...

Another thing I tried was mounting the parent directory of 5 repos into OpenHands; it worked, but polluted the root with .vscode files... It did need extra guidance to understand...