[QA] Staging agent instance dies after 1 hour of inactivity

[joe-rlo]

Environment

hosted-staging (crab shack)

Version / Commit Hash

N/A (hosted-staging)

QA Test Date

2026-04-10

Feature Area

Infrastructure → Agent instance lifecycle / Persistence

Steps to Reproduce

1. Create and start an agent instance on hosted-staging
2. Interact with the agent to confirm it is running
3. Leave the agent idle for approximately 1 hour (no messages sent)
4. Return and attempt to interact with the agent again

Expected Behavior

The agent instance should remain running and responsive even after periods of inactivity. Long-running agents (monitoring bots, scheduled task runners, etc.) need to persist beyond short idle windows.

Actual Behavior

The agent instance at bold-cat-sadip.agents-staging.near.ai shows "ERROR: This agent is not currently running" after approximately 1 hour of inactivity. The staging instance does not persist through idle periods, which makes it impossible to use for any long-running or background task.

Tester confirmed: "Instance dead after 1h of inactivity."

Logs / Screenshots

Agent URL: bold-cat-sadip.agents-staging.near.ai
Error message: "ERROR: This agent is not currently running"
Observed during Item 20 testing by Artem on hosted-staging.

Checklist

• Tested on latest staging or commit
• Steps are reproducible
• Searched for existing issues

[think-in-universe]

This has been fixed in staging. But let's see whether it happens again.

[joe-rlo]

Bug Bash 4/17 — Still reproducing + new details

During today's bug bash, a tester observed the Telegram connection dropping after approximately 5 hours of use:

• Tester had an active Telegram bot connection that was working for several hours
• • After ~5 hours, the connection dropped silently (no error message shown to user)
• • • When attempting to reconnect, the agent asked the user to re-enter the Telegram bot token from scratch rather than using the previously saved token
• • • • This adds a UX friction layer on top of the connection instability — users should not need to re-enter credentials for a reconnection
        This confirms the issue is still active on staging and adds the detail about token re-entry being required on reconnection.

Environment: hosted-staging (crab shack), v0.25.0
Date: 2026-04-17