solpbc.org / solstone
personal memory agent
0
fork

Configure Feed

Select the types of activity you want to include in your feed.

Simplify extraction token settings to fix max_tokens truncation

Overnight schedule extraction was hitting max_tokens limit causing
incomplete JSON responses. Changed extraction hooks to use simpler
settings: 16K output tokens with thinking disabled (was 49K output
+ 24K thinking). Extraction is deterministic markdown-to-JSON
transformation that doesn't benefit from extended reasoning.

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Jer Miller 6df91ca1 6b5f1142

+4 -4
+2 -2
muse/anticipation.py
··· 63 63 contents=contents, 64 64 context=f"agent.{name}.extraction", 65 65 temperature=0.3, 66 - max_output_tokens=8192 * 6, 67 - thinking_budget=8192 * 3, 66 + max_output_tokens=16384, 67 + thinking_budget=0, 68 68 system_instruction=prompt_content.text, 69 69 json_output=True, 70 70 )
+2 -2
muse/occurrence.py
··· 68 68 contents=contents, 69 69 context=f"agent.{name}.extraction", 70 70 temperature=0.3, 71 - max_output_tokens=8192 * 6, 72 - thinking_budget=8192 * 3, 71 + max_output_tokens=16384, 72 + thinking_budget=0, 73 73 system_instruction=prompt_content.text, 74 74 json_output=True, 75 75 )