Agentless solves software issues by first localizing faults hierarchically to files, functions, and edit locations. It then generates multiple candidate patches in diff format and validates them using regression tests and reproduction tests. Unlike agent-based methods, Agentless is simpler, cheaper, and achieves state-of-the-art results on SWE-bench Lite (27.3% solve rate) and SWE-bench Verified (50.8% with Claude 3.5 Sonnet), costing on average $0.34 per issue. It is designed for researchers and developers working on automated bug fixing and software engineering benchmarks.
Pricing model
Free
Starting price
Free (open-source, requires API key for LLM calls)
Agentic power
Autonomous
Context window
—
