openJiuwen presents TTSE agent self-evolution framework with ALFWorld and PinchBench gains
Overview
openJiuwen presents TTSE (Two-Track Self-Evolution), a framework that stores agent experience in two separate types of entries: FACT entries for environment facts and TIP entries for conditional execution tips.
The company reports that on five benchmarks, TTSE reaches up to 98.51% success on ALFWorld with GLM-5 and adds 3.83 points on PinchBench's 147 real-world tasks, according to a QbitAI (量子位) report.
TTSE is available in WorkSwarm 0.2.7 and later. Users turn it on with a TTSE self-evolution toggle in the settings.
Written by AI from the articles below · updated Nov 10, 5:08 AM ET
Check the sources:
Article timeline
The articles in this story. Times are ET.
- QbitAINewsopenJiuwen releases TTSE, a two-track framework for agent self-evolution
AIopenJiuwen presents TTSE (Two-Track Self-Evolution), a framework that separates agent experience into FACT entries for environment facts and TIP entries for conditional execution tips. On five benchmarks, the company reports ALFWorld success up to 98.51% on GLM-5 and a 3.83-point gain on PinchBench's 147 real-world tasks. TTSE is available in WorkSwarm 0.2.7 and later through a settings toggle.
Heat Trend
Not enough continuous observations to show a trend yet.