16:32:35+00:00 [agent] low context window: ollama/qwen3.5:0.8b ctx=16384 (warn<32000) source=model
16:32:36+00:00 [agent] [context-overflow-precheck] sessionKey=agent:main:main provider=ollama/qwen3.5:0.8b route=compact_only estimatedPromptTokens=78 promptBudgetBeforeReserve=1 overflowTokens=77 toolResultReducibleChars=0 reserveTokens=16384 sessionFile=/home/node/.openclaw/agents/main/sessions/da60b568-2654-4d9c-9631-c9fb413bcc7f.jsonl
16:32:36+00:00 [agent] [context-overflow-diag] sessionKey=agent:main:main provider=ollama/qwen3.5:0.8b source=promptError messages=1 sessionFile=/home/node/.openclaw/agents/main/sessions/da60b568-2654-4d9c-9631-c9fb413bcc7f.jsonl diagId=ovf-mnvzfbc6-Gqk76Q compactionAttempts=0 observedTokens=unknown error=Context overflow: prompt too large for the model (precheck).
16:32:36+00:00 [agent] context overflow detected (attempt 1/3); attempting auto-compaction for ollama/qwen3.5:0.8b
16:32:37+00:00 [compaction-safeguard] Compaction safeguard: no real conversation messages to summarize; writing compaction boundary to suppress re-trigger loop.
16:32:37+00:00 [agent] auto-compaction succeeded for ollama/qwen3.5:0.8b; retrying prompt
16:32:37+00:00 [agent] [context-overflow-precheck] sessionKey=agent:main:main provider=ollama/qwen3.5:0.8b route=compact_only estimatedPromptTokens=122 promptBudgetBeforeReserve=1 overflowTokens=121 toolResultReducibleChars=0 reserveTokens=16384 sessionFile=/home/node/.openclaw/agents/main/sessions/da60b568-2654-4d9c-9631-c9fb413bcc7f.jsonl


And the command returns:

```json
{
  "ok": true,
"capability": "model.run",
  "transport": "local",
  "provider": "ollama",
  "model": "qwen3.5:0.8b",
  "attempts": [],
  "outputs": [
    {
      "text": "Context overflow: prompt too large for the model. Try /reset (or /new) to start a fresh session, or use a larger-context model.",
      "mediaUrl": null
    }
  ]
}
## Expected behavior
If `agents.defaults.compaction.reserveTokens=2048` and `reserveTokensFloor=0` are configured and visible via `openclaw config get`, the precheck should use those values instead of `16384`.

At minimum, the runtime should not hardcode or retain the old reserve value after config override + restart.

## Notes

Originally I also hit idle-timeout behavior with larger context settings, but the confirmed issue here is narrower and reproducible: the configured compaction reserve values are not honored in the runtime precheck path.

### Steps to reproduce

Start OpenClaw using a local ollama model on a Raspberry PI.

### Expected behavior

I can use the chat.

### Actual behavior

I get the following error:
Context overflow: prompt too large for the model. Try /reset (or /new) to start a fresh session, or use a larger-context model.

### OpenClaw version

2026.4.10`

### Operating system

Raspberry Pi OS

### Install method

docker

### Model

ollama --> `qwen3.5:0.8b`

### Provider / routing chain

openclaw -> ollama

### Additional provider/model setup details

_No response_

### Logs, screenshots, and evidence

```shell