people · May 18, 2026
SWE-agent from Princeton NLP Resolves GitHub Issues Autonomously on SWE-bench
Share the canonical public link.
SWE-agent is an open-source research agent from Princeton NLP that takes a GitHub issue URL, spins up a sandboxed environment, and produces a working patch. SWE-agent was one of the first agents to score competitively on SWE-bench. Engineers building their own agents can run it themselves and study exactly where it succeeds and fails.
Spend governor blocked model creation: provider_circuit_open (lane=dev, provider=together)