Agent times out after 2 minutes on long-running reasoning tasks #4270

Summary

The agent surfaces a hard 2-minute timeout error ("No response from the agent after 2 minutes. Try again or check your connection.") on tasks that legitimately require longer processing — leaving the user with no response and a reset input field.

Problem

What happened: After sending a message ("Can you summarize RecMem? And understand how it can be used in OpenHuman's codebase?") in Reasoning mode, the agent failed to respond within 2 minutes and triggered a timeout error. The message was sent 7m ago with no response received. Context was 22% (223K/1.0M tokens).

What was expected: For long-running reasoning or research tasks, the agent should either:

• Complete within a longer timeout threshold, or
• Stream partial output to keep the connection alive and prevent a premature timeout.

Steps to reproduce:

1. Start a conversation in Reasoning mode with a substantive research/analysis task.
2. Wait — observe the "No response from the agent after 2 minutes" error appear.

Platform: macOS desktop.

Solution (optional)

• Increase the timeout threshold for Reasoning mode tasks, or make it configurable per mode.
• Implement streaming/heartbeat tokens during long reasoning phases to prevent connection drops.
• Investigate whether the timeout is client-side (UI) or server-side (core RPC), and fix at the right layer.

Acceptance criteria

• Repro gone — Reasoning mode tasks that take > 2 minutes complete successfully without a timeout error.
• Regression safety — Genuine connectivity failures still surface an appropriate error.
• Diff coverage ≥ 80% — Fix PR meets the changed-lines coverage gate.

Related

Related: #4269 (Research agent intermittent hang — may share root cause)