Agentless is an open-source automated software development framework developed by OpenAutoCoder. Unlike complex multi-agent architectures that rely on autonomous exploration and tool-calling loops, Agentless adopts a structured, deterministic pipeline approach to solving software engineering tasks and benchmark problems like SWE-bench. The system operates via modular command-line scripts across discrete phases. It first performs hierarchical localization to identify relevant files, classes, and code lines using repo structure analyzers and search tools. It then prompts supported frontier foundation models (such as OpenAI GPT-4o and Anthropic Claude) to generate candidate patches for the localized code regions. Finally, it validates, filters, and reranks candidate patches using syntax verification and test-suite validation. What sets Agentless apart is its intentional rejection of complex agentic scaffolding. By avoiding open-ended autonomous decision loops, tool hallucination, and runaway token expansion, it achieves high benchmark resolution rates at a fraction of the token cost and execution overhead seen in traditional agentic harnesses. Agentless is open-source software available under an open license on GitHub. Users supply their own LLM API keys, paying standard provider API rates per token consumed.
Hierarchical localization — Pinpoints target files, functions, and lines through repository structural indexing and targeted search. - Deterministic multi-phase pipeline — Replaces unpredictable autonomous agent loops with discrete localization, patch generation, and validation phases. - Candidate patch generation — Generates multiple candidate diffs in parallel for localized problem areas. - Patch validation and reranking — Tests and filters generated candidate solutions using syntax checking, test runner execution, and ranking scripts. - SWE-bench evaluation harness — Provides out-of-the-box support for running, measuring, and analyzing SWE-bench Lite and Verified datasets. - Token and cost efficiency — Eliminates iterative tool-calling overhead, significantly reducing API inference costs compared to interactive agents.
Use cases
Use Case: Automated Benchmark Evaluation Researchers and engineers running SWE-bench or SWE-bench Lite evaluations can use Agentless as a fast, deterministic baseline harness to benchmark foundation models against complex issue-resolution datasets.
Use Case: Batch Issue Resolution Engineering teams can run the pipeline against repositories containing recorded bug reports to generate, test, and rank plausible patch diffs in batch environments without interactive intervention.
Use Case: Program Repair Experimentation Developers experimenting with localized prompt engineering, custom retrieval algorithms, or patch reranking heuristics can isolate and modify individual stages of the modular pipeline without managing dynamic agent state.
Who it is for
Agentless is designed for AI researchers, tool builders, and platform engineers who need a deterministic, cost-effective baseline for automated program repair and benchmark evaluation. It is not intended for developers looking for an interactive conversational terminal copilot or an IDE-integrated code completion assistant.
Install guide
Installation Clone the repository and install the required dependencies in a Python 3.10+ virtual environment: bash git clone https://github.com/OpenAutoCoder/Agentless.git cd Agentless python3 -m venv .venv source .venv/bin/activate pip install -r requirements.txt export PYTHONPATH=$PYTHONPATH:$(pwd)
First Run Set your model provider API credentials as environment variables: bash export OPENAI_API_KEY="your-openai-api-key" export ANTHROPIC_API_KEY="your-anthropic-api-key"
Example Session Run the localization and patch generation stages against a target issue using the Agentless CLI modules: bash # Step 1: Run hierarchical localization python -m agentless.fl.localize \ --repo_path /path/to/target/repo \ --problem_file /path/to/problem_statement.json \ --output_folder ./localization_output # Step 2: Generate candidate patches from localized targets python -m agentless.repair.repair \ --loc_file ./localization_output/loc_results.json \ --output_folder ./repair_output # Step 3: Rerank and filter generated patches python -m agentless.repair.rerank \ --repair_folder ./repair_output \ --output_folder ./final_patch
Continue — Continue is an open-source AI code assistant and agent framework originally developed by the Continue team (continuedev) and recently acquired…
agentdiff — agentdiff is an open-source, Rust-based provenance tracking tool designed to establish Git-native attribution for AI-generated code. Created by Prakhar (codeprakhar25),…
Amp — Amp (developed by AmpCode) is an autonomous coding agent platform and remote development harness designed to execute software engineering tasks…
Cline — Cline is an open-source autonomous coding agent runtime maintained by Cline Bot Inc. and an active open-source community. Available as…
Kiro — Kiro is an agentic engineering platform and CLI coding agent built and operated by AWS. It extends beyond standard code-completion…
OpenHands — OpenHands (formerly OpenDevin) is an open-source autonomous AI software development agent and orchestration platform. It provides an extensible runtime environment…