feat: add /clear endpoint for V1 conversations by MkDev11 · Pull Request #12786 · OpenHands/OpenHands

MkDev11 · 2026-02-06T23:31:52Z

Summary

This PR refactors the /clear command for V1 conversations to use Option 4: New conversation ID, same runtime. Instead of deleting event files (which violated the append-only invariant), /clear now creates a brand new conversation that inherits the parent's sandbox and configuration. The old conversation is preserved and linked via parent_conversation_id.

What changed

Backend (app_conversation_router.py):

Rewrote the clear_conversation endpoint to call start_app_conversation with the current conversation as the parent
The endpoint fully awaits the new conversation's creation before returning, so the frontend can safely redirect to it
Returns the new conversation ID in hex format (no dashes) to match the URL convention used throughout the app

Frontend (chat-interface.tsx, v1-conversation-service.api.ts):

Added a clearConversation static method to V1ConversationService
When the user types /clear in a V1 conversation, the frontend calls the API, shows a success toast, and navigates to the new conversation using React Router

Infrastructure (deployment-related, not core feature):

docker_sandbox_service.py: CORS origins are now passed as indexed environment variables (OH_ALLOW_CORS_ORIGINS_0, _1, etc.) to agent server containers, supporting multiple allowed origins
websocket-url.ts: When accessing the frontend from an external IP, WebSocket URLs that point to localhost are rewritten to use the browser's hostname
middleware.py: Falls back to allowing all origins when no specific CORS origins are configured

Tests:

Updated existing unit tests for the new response format (hex IDs, app_conversation_id)
All 11 tests pass

How it works

User types /clear in a V1 conversation
Frontend calls POST /api/v1/app-conversations/{id}/clear
Backend creates a new conversation with parent_conversation_id set to the old one — same sandbox, same agent config, same LLM model
Backend waits for the conversation to be fully saved to the database, then returns the new conversation ID
Frontend navigates to /conversations/{new_id} — fresh chat history, runtime state preserved

Demo Screenshots/Videos

Loom recording showing /clear working end-to-end

Change Type

Checklist

I have read and reviewed the code and I understand what the code is doing.
I have tested the code to the best of my ability and ensured it works as expected.

Fixes

Resolves #12564

Release Notes

Include this change in the Release Notes.

Added /clear command for V1 conversations. Typing /clear in the chat creates a new conversation that inherits the current sandbox and configuration, giving you a fresh chat history while keeping your runtime state (files, packages, environment) intact.

MkDev11 · 2026-02-06T23:32:23Z

@enyst Just opened the PR for v1

all-hands-bot

This PR adds a useful clear endpoint, but there are some important security and error handling concerns that should be addressed before merging.

MkDev11 · 2026-02-06T23:45:50Z

This PR adds a useful clear endpoint, but there are some important security and error handling concerns that should be addressed before merging.

Fixed all!

all-hands-bot

Overall solid implementation of the /clear command. Found several security and design concerns that should be addressed, particularly around CORS configuration and the contradictory clear_events implementation.

MkDev11 · 2026-02-09T12:38:26Z

@enyst hope you had a good weekend! could you please review the changes and let me know your feedback?

enyst

Thank you for the work on this, @MkDev11 ! I started a discussion on it on slack.

hieptl

Hello @MkDev11,

Thank you for creating this pull request — much appreciated. 🙏

I’ve left a few comments on the PR for your reference. Overall, I believe we should avoid creating a new conversation and instead clear the events of the current conversation.

@tofarr, I would greatly appreciate your feedback and guidance on this approach.

Additionally, could you please add some front-end tests to help cover the changes introduced here?

Thank you very much, and I appreciate your work on this! 🙏

MkDev11 · 2026-02-09T16:00:33Z

Hello @MkDev11,

Thank you for creating this pull request — much appreciated. 🙏

I’ve left a few comments on the PR for your reference. Overall, I believe we should avoid creating a new conversation and instead clear the events of the current conversation.

@tofarr, I would greatly appreciate your feedback and guidance on this approach.

Additionally, could you please add some front-end tests to help cover the changes introduced here?

Thank you very much, and I appreciate your work on this! 🙏

The current approach (creating a new conversation with parent_conversation_id) was chosen to preserve the append-only event stream invariant. Deleting events from the agent-server side would break event replay, audit trails, and any downstream consumers that rely on event IDs being immutable. @tofarr — would appreciate your guidance on whether clearing events in-place on the agent-server side is preferred instead.

MkDev11 · 2026-02-09T16:01:44Z

@hieptl could you please have a look at the changes again?

MkDev11 · 2026-02-18T09:42:59Z

@tofarr thanks for your feedback. could you please review again?

tofarr

This is some great work - thank you for taking this on. I've left a few comments, and will review again if you have any questions or changes.

MkDev11 · 2026-02-18T10:21:02Z

@tofarr fixed all the comments.

hieptl · 2026-02-25T16:56:22Z

Hello @MkDev11,

Could you please add tests to cover your front-end changes when you have a chance?

In addition, I ran the pull request locally and noticed a few issues.

Issue 1:
After using the /clear command, the previous conversations are still displayed in the conversation list. Please refer to the video below for more details.

issue1.mov

Issue 2:
It’s still possible to access the old conversations after using /clear. It appears that these conversations are not deleted and remain accessible. Please see the video below for reference.

issue2.mov

Issue 3:

Step 1: Create a new conversation.
Step 2: Send a simple message (e.g., “Hello”).
Step 3: Use the /clear command.
Step 4: Send another message in the new conversation (e.g., “Who are you?”).
Step 5: While the new conversation’s title is updated, the title shown in the conversation list is not updated accordingly.

Please refer to the video below for more information.

issue3.mov

Issue 4:
There appears to be a noticeable delay after sending the /clear command. In cases where the WebSocket connection is interrupted or there are network issues, this could lead to a suboptimal user experience. The video below provides additional context.

issue5.mov

Thank you very much! 🙏

MkDev11 · 2026-02-26T02:02:33Z

@hieptl fixed the issues you mentioned. please review again

hieptl · 2026-03-02T12:01:44Z

Hello @MkDev11,

I noticed the issue below while running the code locally. Please feel free to refer to the video below for additional details and context.

issue.mov

Thank you very much! 🙏

enyst · 2026-03-09T09:31:39Z

I think maybe it makes a lot of sense, and it would be expected, that the "application control commands" executed on slash /command would be in the menu.

By control commands, I mean those that are not skills. This one is not a skill, it's pure python implementation of a control command. (Of course, we include skills too, those look like slash commands and will be interpreted by the agent as commands for itself; I'm just saying this one is not an agent thing, it's an application thing)

Though I'd add a thought, they make sense to me at the start of the line... like a CLI command.

Could be in a future PR, I suppose? I'm not sure if other control commands are in the slash menu currently.

However, since these conversations share the same runtime, clicking “Close Conversation (Stop Runtime)” will stop the runtime for all of them. I’m not certain whether this behavior is intended.

You're right, that's clearly the case...

We could do as you suggest:

Conversation cleared. Starting a new conversation in the same sandbox. These conversations will share the same runtime.

Would it make sense to rename the command from /clear to /new? This might better reflect the behavior of starting a new conversation.

Yes. I believe we should, it would be inline with what it does, and easier to understand for users coming from other agentic applications.

MkDev11 · 2026-03-10T15:13:12Z

@tofarr @hieptl do you want me to rename /clear->/new, update toast message, add cursor-not-allowed, add to slash menu?

hieptl · 2026-03-12T08:51:46Z

Hello @MkDev11,

As Engel mentioned:

Adding this command to the slash command menu can be included in a future pull request.
The toast message needs to be updated.
The command should be changed from /clear to /new.

Additionally, based on the two issues I noted in my previous comment, both will need to be addressed.

Thank you very much! 🙏

…-allowed, fix cross-conversation change sync - Rename /clear command to /new in chat-interface.tsx - Update i18n translations to reflect /new command wording - Update toast success message to clarify shared runtime - Add cursor-not-allowed and opacity-50 to disabled chat input - Add refetchOnMount: 'always' to useUnifiedGetGitChanges so navigating between shared-sandbox conversations refetches git changes

MkDev11 · 2026-03-13T03:24:57Z

@tofarr @hieptl do you want me to rename /clear->/new, update toast message, add cursor-not-allowed, add to slash menu?

I refactored the implementation.

…mmand

hieptl · 2026-03-13T12:59:09Z

Hello @MkDev11,

While testing your code locally, I came across a few issues that might be worth reviewing.

Issue 1:

Step 1: Create a new conversation.
Step 2: Send a simple message, such as “Hello, World.”
Step 3: Send /new to create another conversation.
Step 4: In the conversation panel (conversation list), switch to the newly created conversation.
Step 5: Send another message, such as “Hi.”
Step 6: Send /new again to create a third conversation. At this point, all three conversations share the same sandbox.
Step 7: Navigate back to the home page and delete one conversation. The remaining two conversations then appear to be archived.

For additional context, please refer to the video below.

issue1.mov

Issue 2:
When a new conversation is created and begins to start up, the chat input becomes disabled. This appears to be a regression. Ideally, the chat input should remain enabled so that users can type their messages while the conversation is initializing.

In one of my earlier comments, I suggested disabling the chat input only during the brief moment when a new conversation is being created after the user enters /new. However, it seems that the current implementation disables the chat input more broadly than intended.

For additional context, please refer to the video below.

issue2.mov

Thank you very much! 🙏

…rsations When deleting a conversation that shares a sandbox with other conversations (created via /new), skip the DELETE /api/conversations call to the agent server. This prevents destabilizing the shared runtime, which would cause remaining conversations to appear as ARCHIVED. - Add skip_agent_server_delete param to delete_app_conversation - Check count_conversations_by_sandbox_id before deletion - Pass skip flag when sandbox is shared (count > 1) - Re-check remaining count after deletion before sandbox cleanup

Thread a separate inputLocked prop through the component chain so that ChatInputField is only locked (contentEditable=false, cursor-not-allowed) during /new command processing, not during AgentState.LOADING. - Add inputLocked prop to CustomChatInput, ChatInputContainer, ChatInputRow - ChatInputRow passes inputLocked (not disabled) to ChatInputField - disabled still controls send button, file button, and Enter submission - ChatInputField only gets disabled=true during /new (isClearingConversation) This fixes the regression where typing was blocked during conversation startup, while preserving cursor-not-allowed feedback during /new.

…mmand # Conflicts: # frontend/src/components/features/chat/interactive-chat-box.tsx # frontend/src/i18n/declaration.ts # frontend/src/i18n/translation.json # openhands/server/routes/manage_conversations.py

MkDev11 · 2026-03-18T14:50:37Z

@hieptl I fixed the issues you mentioend. can you review the change?

…by_sandbox_id and skip_agent_server_delete

…ther skipped tests)

MkDev11 · 2026-03-19T12:45:52Z

@hieptl do you have any feedbacK?

…mand

MkDev11 · 2026-03-19T13:11:48Z

@hieptl just fixed the lint error

hieptl

Hello @MkDev11,

For the front-end changes, it looks like the current updates focus on adding tests for use-clear-conversation. It would be helpful to also include tests that cover the changes made in the component files. There’s no need to add tests specifically for styling.

Thank you! 🙏

- Fix flaky WebSocket onClose test (reset spy after connection instead of skipping) - Rename inputLocked → isNewConversationPending for clarity - Use cn() instead of template string in chat-input-field.tsx - Rename use-clear-conversation → use-new-conversation-command (file, hook, tests) - Move get_global_config import to top of middleware.py - Add component tests for disabled input field behavior

MkDev11 · 2026-03-19T14:00:20Z

all of your comments are addressed, could you please review again?

The changes look good to me. Thank you very much!

hieptl

Thank you! 🙏

MkDev11 · 2026-03-19T14:14:35Z

thanks for your help!

MkDev11 requested review from hieptl, malhotra5 and tofarr as code owners February 6, 2026 23:31

all-hands-bot reviewed Feb 6, 2026

View reviewed changes

MkDev11 requested a review from amanape as a code owner February 9, 2026 00:48

MkDev11 requested a review from all-hands-bot February 9, 2026 05:50

all-hands-bot reviewed Feb 9, 2026

View reviewed changes

enyst reviewed Feb 9, 2026

View reviewed changes

hieptl previously requested changes Feb 9, 2026

View reviewed changes

MkDev11 requested review from enyst and hieptl February 9, 2026 16:10

MkDev11 mentioned this pull request Feb 14, 2026

[Bug]: cannot use web interface from different machine #12679

Closed

1 task

tofarr reviewed Feb 18, 2026

View reviewed changes

Comment thread openhands/app_server/sandbox/docker_sandbox_service.py Outdated

tofarr reviewed Feb 18, 2026

View reviewed changes

Comment thread openhands/app_server/sandbox/docker_sandbox_service.py Outdated

Comment thread openhands/app_server/app_conversation/app_conversation_router.py

Comment thread openhands/app_server/app_conversation/app_conversation_router.py Outdated

MkDev11 requested a review from tofarr February 18, 2026 10:20

tofarr reviewed Feb 18, 2026

View reviewed changes

Comment thread openhands/app_server/app_conversation/app_conversation_router.py Outdated

tofarr reviewed Feb 18, 2026

View reviewed changes

Comment thread openhands/app_server/app_conversation/app_conversation_router.py Outdated

MkDev11 force-pushed the feature/v1-clear-command branch from 094642a to 84f2393 Compare February 19, 2026 14:28

MkDev11 requested a review from tofarr February 19, 2026 14:44

MkDev11 force-pushed the feature/v1-clear-command branch from d0c9c53 to b5f9831 Compare March 2, 2026 13:43

Merge branch 'main' into feature/v1-clear-command

7e478c0

Merge remote-tracking branch 'upstream/main' into feature/v1-clear-co…

d5483f8

…mmand

MkDev11 added 3 commits March 18, 2026 12:29

Merge remote-tracking branch 'upstream/main' into feature/v1-clear-co…

6ae4f6e

…mmand # Conflicts: # frontend/src/components/features/chat/interactive-chat-box.tsx # frontend/src/i18n/declaration.ts # frontend/src/i18n/translation.json # openhands/server/routes/manage_conversations.py

MkDev11 added 2 commits March 18, 2026 15:58

fix: update delete_v1_conversation tests to mock count_conversations_…

5230e06

…by_sandbox_id and skip_agent_server_delete

test: skip flaky WebSocket onClose test (same MSW race condition as o…

790dd3c

…ther skipped tests)

hieptl and others added 2 commits March 19, 2026 19:50

erge remote-tracking branch 'upstream/main' into feature/v1-clear-com…

e2b2483

…mand

fix: add missing Catalan (ca) translations for CONVERSATION$CLEAR_* keys

952cf2f

hieptl reviewed Mar 19, 2026

View reviewed changes

MkDev11 added 2 commits March 19, 2026 14:34

fix: update test patch targets for middleware.py top-level import

eb657df

MkDev11 requested a review from hieptl March 19, 2026 13:59

hieptl approved these changes Mar 19, 2026

View reviewed changes

hieptl merged commit 0ec962e into OpenHands:main Mar 19, 2026
43 checks passed

jpelletier1 mentioned this pull request Mar 26, 2026

[Feature]: Add /new to the GUI slash command menu #13596

Closed

mamoodi added the release:cloud-1.18.0 Released in Cloud 1.18.0 label Mar 26, 2026 — with OpenHands AI

1Ckpwee mentioned this pull request Mar 28, 2026

fix: add /new to slash command autocomplete menu #13641

Closed

3 tasks

Conversation

MkDev11 commented Feb 6, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

What changed

How it works

Demo Screenshots/Videos

Change Type

Checklist

Fixes

Release Notes

Uh oh!

MkDev11 commented Feb 6, 2026

Uh oh!

all-hands-bot left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

MkDev11 commented Feb 6, 2026

Uh oh!

all-hands-bot left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

MkDev11 commented Feb 9, 2026

Uh oh!

enyst left a comment

Choose a reason for hiding this comment

Uh oh!

hieptl left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

MkDev11 commented Feb 9, 2026

Uh oh!

MkDev11 commented Feb 9, 2026

Uh oh!

Uh oh!

MkDev11 commented Feb 18, 2026

Uh oh!

tofarr left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

MkDev11 commented Feb 18, 2026

Uh oh!

Uh oh!

Uh oh!

hieptl commented Feb 25, 2026

Uh oh!

MkDev11 commented Feb 26, 2026

Uh oh!

hieptl commented Mar 2, 2026

Uh oh!

enyst commented Mar 9, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

MkDev11 commented Mar 10, 2026

Uh oh!

hieptl commented Mar 12, 2026

MkDev11 commented Feb 6, 2026 •

edited

Loading

hieptl left a comment •

edited

Loading

enyst commented Mar 9, 2026 •

edited

Loading