Description

When an Ollama model (specifically ollama/glm-5:cloud in my case) hits its internal context limit, OpenClaw shows a cryptic error message in the TUI instead of handling it gracefully:

run error: Ollama API error 400: {"StatusCode":400,"Status":"400 Bad Request","error":"prompt too long; exceeded max context length by 4 tokens"}
connected | error

Expected Behavior

1. Proactive prevention: Auto-compact should trigger before hitting the limit (currently waits for idle + 60% threshold, but Ollama's internal limit can be hit before OpenClaw's tracking indicates it's time)
2. Graceful degradation: If overflow occurs, show a helpful message like "Context limit reached. Starting fresh conversation..." and auto-compact/reset
3. Recovery: Don't leave the session in an "error" state - either auto-compact and retry, or guide the user to start fresh

Current Behavior

• Shows cryptic error message
• Session appears "broken" until user manually starts new conversation
• Auto-compact doesn't trigger because it waits for idle state

Environment

• OpenClaw: latest
• Model: ollama/glm-5:cloud (Ollama Cloud)
• Surface: openclaw-tui (gateway-client)

Root Cause

1. OpenClaw's token tracking may not match Ollama's actual context usage
2. Auto-compact is reactive (waits for idle) rather than proactive
3. No graceful handling when the 400 error occurs - just surfaces the raw API error

Related Issues

• Feature: Trigger model fallback on context length exceeded #9986 - Feature: Trigger model fallback on context length exceeded
• [Feature] Automatic session compaction and context length management #31787 - Automatic session compaction and context length management
• Feature: Soft context reset / compaction command #28233 - Soft context reset / compaction command

Suggested Fix

1. Add preflight check before sending to Ollama API
2. If context would exceed limit, auto-trigger /compact and retry
3. If still over limit, show helpful message and suggest /new
4. Improve error message in TUI to be user-friendly