What version of the IDE extension are you using?
JetBrains AI Assistant plugin: 253.30387.186
What subscription do you have?
ChatGPT Plus (paid)
Which IDE are you using?
IntelliJ IDEA 2025.3.2 (Build IU-253.30387.90) (Windows)
What platform is your computer?
Microsoft Windows NT 10.0.26200.0 x64
What issue are you seeing?
After the IDE integration started offering gpt-5.3-codex models, using AI Chat (Codex via ACP) becomes extremely laggy / effectively unusable.
After a few prompts with gpt-5.3-codex, the IDE enters a broken state: even switching back to gpt-5.2-codex remains unusable, because the ACP agent repeatedly attempts to session/resume a persisted session and fails with an internal error:
- JSON-RPC -32603 Internal error
- details: "no rollout found for thread id "
- This triggers a loop: start ACP process → fail resume → stop server → kill process tree → restart.
- The loop correlates with severe UI lag in the AI Chat tool window (almost unusable).
Error excerpt:
`method: "session/resume"
params: {
sessionId: "<SESSION_ID_REDACTED>",
cwd: "<PROJECT_PATH_REDACTED>",
mcpServers: [],
}
code: -32603,
message: "Internal error",
data: {
details: "no rollout found for thread id <SESSION_ID_REDACTED>"
}
Failed to initialize ACP session ... Internal error
... Killing process tree ...
`
What steps can reproduce the bug?
- Open IntelliJ IDEA project.
- Open AI Chat (JetBrains AI Assistant) and select Codex / ACP integration.
- Select gpt-5.3-codex (e.g., medium) and send a few prompts (any normal chat prompts).
- Observe: AI Chat becomes very laggy; soon it becomes nearly unusable.
- Switch model back to gpt-5.2-codex and try again.
- Observe: still laggy/unusable; ACP keeps retrying session/resume for the same sessionId and fails with “no rollout found…”, restarting the process repeatedly.
Notes: once triggered, the issue persists across attempts because the IDE keeps trying to resume the same session/thread.
What is the expected behavior?
- AI Chat should remain responsive/usable after switching to gpt-5.3-codex, and switching back to gpt-5.2-codex should recover if needed.
- If a persisted thread/session cannot be resumed (state missing/expired), the integration should gracefully start a new session (or return a recoverable error), instead of entering a restart loop with Internal error.
Additional information
- Regression: before gpt-5.3-codex became available in the IDE integration, gpt-5.2-codex worked normally.
- Codex CLI usage is still workable; the problem seems tied to IDE ACP integration + session resume state.
- Attaching acp.log with timestamps around the failure helps (shows repeated session/resume + no rollout found + process kill/restart loop).
acp.log
What version of the IDE extension are you using?
JetBrains AI Assistant plugin: 253.30387.186
What subscription do you have?
ChatGPT Plus (paid)
Which IDE are you using?
IntelliJ IDEA 2025.3.2 (Build IU-253.30387.90) (Windows)
What platform is your computer?
Microsoft Windows NT 10.0.26200.0 x64
What issue are you seeing?
After the IDE integration started offering gpt-5.3-codex models, using AI Chat (Codex via ACP) becomes extremely laggy / effectively unusable.
After a few prompts with gpt-5.3-codex, the IDE enters a broken state: even switching back to gpt-5.2-codex remains unusable, because the ACP agent repeatedly attempts to session/resume a persisted session and fails with an internal error:
Error excerpt:
`method: "session/resume"
params: {
sessionId: "<SESSION_ID_REDACTED>",
cwd: "<PROJECT_PATH_REDACTED>",
mcpServers: [],
}
code: -32603,
message: "Internal error",
data: {
details: "no rollout found for thread id <SESSION_ID_REDACTED>"
}
Failed to initialize ACP session ... Internal error
... Killing process tree ...
`
What steps can reproduce the bug?
Notes: once triggered, the issue persists across attempts because the IDE keeps trying to resume the same session/thread.
What is the expected behavior?
Additional information
acp.log