https://docs.all-hands.dev/ logo
Join Slack
Powered by
Plane mode in agent-canvas: <github.com/OpenHands/...
# proj-agent-canvas
m
Plane mode in agent-canvas: github.com/OpenHands/OpenHands/pull/16178 @hundreds-twilight-32833 If you have a moment: can you also test it? I re-tested it and it seems to work as the one in cloud.
👀 1
h
A couple of things on first run through: • Plan mode stays active once enabled even when switching from a conversation where it is active to http://localhost:8000/conversations • We don't have the code/plan toggle like in the old UI so you need to be in a conversation to get to it. Then I can't seem to turn it off. I think a good solution here might be to add a 
/plan
slash command that can enable it at any time. Then either 
/code
option to turn it off or add a UI (Turn of Plan Mode x) to the chat. That might be a nice subtle way to make it flexible. Unless I'm missing something that should be obvious on enabling/disabling?
m
localhost/conversations
I cannot click the link 🙂
...here might be to add a 
/plan
slash command that can enable it at any time. Then either 
/code
...
This should not be difficult to implement Thanks
h
I can show a video. I wanted to be specific that it is the 
/conversations
route after clicking new chat
m
Ah ok 👍
Ready for another pass
🙌 1
😉
h
Whenever I try and use the plan mode my LLMs can't run. Also some weirdness with how 
/plan
kicks things off. I don't want this to become too difficult of an implementation so if you want to shelve this for now we can just come back to the planner mode implementation later. I think the other items we have in the cycle or higher priority. Video for context
Screen Recording 2026-08-03 at 9.42.58 AM.mov
m
ah ok. the problem is with the command. You should just use /plan withouth giving nothing. But this can be changed easily
h
I used 
/plan
and 
/code
inside a conversation and that does work well
m
@hundreds-twilight-32833
👀 1
h
I missed the video earlier, thank you for tagging me! Checking it out now
There must be something different when using a ChatGPT subscription for models. When I use the plan mode my chatGPT LLMs just don't work. I get:
Copy code
litellm.BadRequestError: OpenAIException - {
  "error": {
    "message": "You have insufficient permissions for this operation. Missing scopes: api.responses.write. Check that you have the correct role in your organization (Reader, Writer, Owner) and project (Member, Owner), and if you're using a restricted API key, that it has the necessary scopes.",
    "type": "invalid_request_error",
    "param": null,
    "code": null
  }
}
m
mmmh... gpt has also a /plan command
perhaps there are some conflict with that
h
So I have an agent profile with ChatGPT 5.6 Terra as the default. When I use minimax-m2.7 I still get that same OpenAIException
is the plan mode using the agent profile default somehow?
m
ok weird... it has nothing to do with the profile. It is just intercepting the message before you send it to the model
h
hmm
I'll do a quick restart of things and refresh
👍 1
m
So the problem with the switch is on me... the problem with the subscription seems unrelated
h
I can't seem to get it work on my end 😞 . Maybe because I'm using a bunch of chatgpt models. I'm also wondering if approach with 
/code
and 
/plan
is correct with how its kind of brittle. Here is what I'm seeing even with the change you just pushed up
Screen Recording 2026-08-04 at 4.35.32 PM.mov
2 Views
Open in Slack
PreviousNext
Powered by