Today for AI
HOT RADAR
agentHEAT 13.1°

AI CLUSTERED EVENT · 10/11/2026

Meta's GitSwarm: Multi-Agent System Leverages Shared Git History to Reuse Failed Attempts for Program Reconstruction

1 reports archived1 independent sourcesupdated 10/11/2026, 18:32:55
Synthesis & Latest Updates
1 Sources Cross-Validated

Meta introduces GitSwarm, a framework where multiple homogeneous agents collaborate via a shared Git repository to reuse historical attempts, including failures. Without a central scheduler, agents autonomously read past commits and track dependencies, achieving 79.4% accuracy on ProgramBench—significantly outperforming single-agent continuous execution (65.1%)—demonstrating the critical value of preserving 'failed states' for enhancing agent collaboration efficiency.

LATEST/Meta introduces GitSwarm, a framework where multiple homogeneous agents collaborate via a shared Git repository to reuse historical attempts, including failures. Without a central scheduler, agents autonomously read past commits and track dependencies, achieving 79.4% accuracy on ProgramBench—significantly outperforming single-agent continuous execution (65.1%)—demonstrating the critical value of preserving 'failed states' for enhancing agent collaboration efficiency.

HEAT TRENDHourly heat curve

12 fully observed hours
Now
13.1
Peak
17.320:00
24h change
–

The line compares only sources observed throughout; gap hours are interpolated to keep the trend continuous.

TIMELINECoverage timeline

Total 1 reports · Latest first
  1. Rohan Paul (@rohanpaul_ai)T2·78 pts
    • GitSwarm uses a shared Git repository as long-term memory and collaboration space for agents, preventing the loss of partial results common in traditional single-run methods.
    • Experiments show subsequent agents effectively reuse 94.7% of historical contributions (including failed attempts), proving that failure paths contain valuable exploration information.
    • Under equivalent compute budgets, the multi-agent parallel collaboration mode outperforms single-agent continuous iteration by approximately 14 percentage points on program reconstruction tasks.