(placeholder) SDK flexibility

[enyst]

I'd like to use this issue to list the relevant issues on SDK flexibility, so we can maybe look at a larger picture on SDK architecture and behavior. Please let me know your thoughts, questions, roasting, as it may be. 😅

SDK flexibility

• Proposal: agent as composed of immutable instances #1451
• PR: refactor: remove reconciliation methods, use runtime agent directly #1542
  • a bunch of issues and PRs, in SDK and CLI repos, at least, are related to it
  • I'm not sure there are on Web UI repo, but there will be
• root issue: Multiple LLM working in sync  OpenHands#2075

It allows to switch Agent's components, at restore time and at runtime, without workarounds. We may or may not want all of them, which is a separate question. I think we know we want the most important of them: LLM switch.

This also unlocks the ability for seamless passage from remote to local and back (conversation teleport).

LLM Switch

• Support switching LLM models in the middle of conversation #416
• in SDK:
  • (experimental) Agent sdk 26 llm switch experiment enyst/agent-sdk#17
  • based on: feat(llm): add LLM profiles #795
  • Switch LLM profiles #1049
  • [agent-sdk-j2s] agent-server runtime LLM switching #1544
• in VSCode Client:
  • already implemented in https://github.com/enyst/OpenHands-Tab
  • Remote LLM switching via agent-server /llm enyst/OpenHands-Tab#625

(Note: LLM switch = LLM profile switch is a simplifying assumption that works)

Conversation Teleport

• [Feature] Trajectory Replay OpenHands#6049
• feat: Delegate conversation to cloud OpenHands-CLI#230
• [Feature] Move/Teleport In-Progress Tasks Between CLI and Cloud All-Hands-AI/OpenHands-Cloud#248
• an experimental, simplified implementation is already in OpenHands-Tab.