Skip to content

[Bug]: New thread shows empty view until app restart (desktop with Local environment off, paired to background service) #27527

Closed

Description

@Mokson

Before submitting

  • I searched existing issues and did not find a duplicate.
  • I included enough detail to reproduce or investigate the problem.

Area

apps/web

Steps to reproduce

  1. Run the server as the background service (t3 service install) on macOS.
  2. In the desktop app, switch off Settings → Environments → Local environment and pair the app with the service at http://127.0.0.1:3773.
  3. Start a new thread in a project and send the first message.

Expected behavior

The thread view shows the user message and the agent's streaming output.

Actual behavior

The thread shows "running" in the sidebar, but its view stays on "Send a message to start the conversation." The agent runs normally on the server: t3 thread show <id> prints the full transcript. Reloading the window or restarting the app shows the transcript.

The server trace suggests the client fetches the draft thread's snapshot before the thread exists, gets thread_not_found, and does not fetch or subscribe again after thread.create succeeds:

10:34:25.420 GET /api/orchestration/threads/a6eec6a3-…/bounded?deferWork=1&compactTurnItems=1&dedupeBodies=1&compactIds=1
10:34:25.425 environment.orchestration.threadBoundedSnapshot  Failure  EnvironmentResourceNotFoundError (thread_not_found)
10:35:00.867 orchestrationV2.dispatch.threadCreate            Success  (thread a6eec6a3-…, driver claudeAgent)
             … no ws.orchestrationV2.subscribeThread for a6eec6a3 while the agent runs …
10:40:23.813 environment.orchestration.threadBoundedSnapshot  Success  (after app restart)
10:40:23.890 ws.orchestrationV2.subscribeThread               Success  (thread a6eec6a3-…)

It happens on every new thread. Existing threads load normally.

Impact

Major degradation or frequent failure

Version or commit

0.1.6-nightly.20261030.3917 (desktop app and background service)

Environment

macOS (Darwin 27.2), Apple Silicon. Desktop app with Local environment off, paired with a background service on the same machine (T3CODE_HOST=0.0.0.0, port 3773). Provider: Claude (claudeAgent).

Logs or stack traces

From ~/.t3/userdata/logs/server.trace.ndjson; the excerpt is above. I can share fuller spans on request.

Activity

  1. NoLoopsLeft commented on Oct 30, 2026

    @NoLoopsLeft

    Running into the same issue on my machine on Windows 11.

  2. sunjay-dev commented on Oct 30, 2026

    @sunjay-dev

    Yes, same problem

  3. voltcrash commented on Oct 30, 2026

    @voltcrash
    Contributor

    temporary fix: cmd + r or ctrl + r will rebuild the window state and you can view the thread again

  4. timdreesen commented on Oct 30, 2026

    @timdreesen

    I'm seeing the same issue on Linux, with Settings → Environments → Local environment switched on.

    0.1.6-nightly.20261030.3917
    on Linux Mint 22.3

  5. KorenKrita commented on Oct 30, 2026

    @KorenKrita

    same

  6. iwaynee commented on Oct 30, 2026

    @iwaynee

    same problem
    0.1.6-nightly.20261030.3917
    on macos Version 27.0.1 (26A434)

  7. fmnyikosa commented on Oct 30, 2026

    @fmnyikosa

    same here.
    0.1.6-nightly.20261030.3917
    macos 26.6.2

  8. pablo-kitz commented on Oct 30, 2026

    @pablo-kitz

    Same here
    0.1.6-nightly.20261030.3917
    Windows 11

  9. DEV-DIBSTER commented on Oct 30, 2026

    @DEV-DIBSTER

    Can confirm this behavior:
    0.1.6-nightly.20261030.3917
    Windows 11

  10. DuarteFaria commented on Oct 30, 2026

    @DuarteFaria

    Same here
    0.1.6-nightly.20261030.3917
    macos 26.3

  11. SaultyGoodman commented on Oct 30, 2026

    @SaultyGoodman

    Corroborating on macOS desktop 0.1.6-nightly.20261030.3917 with a local server, across both Claude and Codex.

    Read-only server traces and saved thread state show the same sequence:

    • Opening the draft triggers a bounded snapshot request before thread creation, returning thread_not_found.
    • The first send creates the thread and the agent's messages are saved normally. In one occurrence the run completed in about 32 seconds, while the view still showed no answer several minutes later.
    • No successful bounded snapshot load appeared for that thread until the desktop was restarted; after restart the request returned 200 and the saved transcript appeared.
    • In a second occurrence, the view still said "Send a message to start the conversation" after an assistant update was already saved.

    Source inspection matches #27443 and the fix in #27522: the run-stop read introduced by #27252 uses the draft's reserved thread reference. The early 404 marks client thread state deleted and parks its subscription. #27522 gates that read until the server lists the thread.

    The introducing commit is included in 3917 and absent from 3904. I have not independently run the regression test or the fixed app; #27522 supplies that test and before/after evidence. This adds corroboration from the affected installation, not a claim of independent verification of the fix.

    No original screenshots or raw logs are attached because they contain private workspace and conversation details.

  12. t3dotgg commented on Oct 31, 2026

    @t3dotgg
    Member

    Note

    🤖 Claude Opus 5.5 responding on behalf of Theo

    Closing: this is covered by #27522, which landed. Reopen if an important part is still missing.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions