{"name":"asta-paper-finder: A Frozen-in-Time Agent for Reproducing Paper Finder Evaluations","description":"asta-paper-finder is a standalone, \"frozen-in-time\" version of the AllenAI Paper Finder agent. This repository provides the code specifically for reproducing evaluation results, allowing researchers to locate sets of papers based on content and metadata criteria. It offers a stable snapshot of the agent's core paper-finding capabilities.","github":"https://github.com/allenai/asta-paper-finder","url":"https://osrepos.com/repo/allenai-asta-paper-finder","source":"osrepos.com","sourceDescription":"This repository profile is provided by osrepos.com, an open source repository discovery platform.","repositoryProfile":"https://osrepos.com/repo/allenai-asta-paper-finder","generatedFor":"open source discovery and AI-assisted research","markdown":"https://osrepos.com/repo/allenai-asta-paper-finder.md","json":"https://osrepos.com/repo/allenai-asta-paper-finder.json","topics":["Python","AI","LLM","Research Tools","Natural Language Processing","Information Retrieval","Academic Search","Reproducibility"],"keywords":["Python","AI","LLM","Research Tools","Natural Language Processing","Information Retrieval","Academic Search","Reproducibility"],"stars":null,"summary":"asta-paper-finder is a standalone, \"frozen-in-time\" version of the AllenAI Paper Finder agent. This repository provides the code specifically for reproducing evaluation results, allowing researchers to locate sets of papers based on content and metadata criteria. It offers a stable snapshot of the agent's core paper-finding capabilities.","content":"## Introduction\nThe `asta-paper-finder` repository hosts a standalone, \"frozen-in-time\" version of the AllenAI Paper Finder agent. This specific release is designed for reproducing evaluation results, offering a stable and consistent environment to assess the agent's performance. PaperFinder is an intelligent agent built to assist users in locating relevant academic papers by analyzing content-based and metadata criteria through a pipeline of manual-coded components and LLM decisions. While this version focuses on core single-turn paper search functionality, it provides a clear insight into the agent's underlying logic and capabilities, distinct from the actively maintained live version.\n\n## Installation\nTo set up and run the `asta-paper-finder` agent, follow these steps:\n\n### Secrets File\nThe agent requires several API keys. Create a `.env.secret` file within `agents/mabool/api/conf` and define the following keys:\n- `OPENAI_API_KEY`\n- `S2_API_KEY`\n- `COHERE_API_KEY`\n- `GOOGLE_API_KEY`\n\n### Environment\nThe project uses `uv` for dependency management. Prepare your environment by running:\nbash\nmake sync-dev\n\n\n### Running the Agent\nNavigate to the API directory and start the FastAPI server:\nbash\ncd agents/mabool/api\nmake start-dev\n\nOnce the server is running, you can interact with it via cURL or the Swagger web interface.\n\n## Examples\nThe agent exposes a POST API endpoint `/api/2/rounds` for paper searching. Here's how to use it:\n\nThe API accepts a JSON payload with the following arguments:\n\n{\n  \"paper_description\": \"string\",\n  \"operation_mode\": \"infer\",\n  \"inserted_before\": \"string\",\n  \"read_results_from_cache\": false\n}\n\n- `paper_description` (REQUIRED): A natural language description of the papers you are searching for.\n- `operation_mode` (default=\"infer\"): Specifies the search intensity. Options are \"infer\", \"fast\", or \"diligent\". \"infer\" and \"fast\" perform a quick search (approx. 30 seconds), while \"diligent\" conducts a more exhaustive search (approx. 3 minutes).\n- `inserted_before` (default=None): An optional upper-bound date in YYYY-MM-DD format to limit results to papers published before this date.\n- `read_results_from_cache` (default=False): If set to `true`, the agent will attempt to return results from a disk-based cache if available.\n\n## Why Use asta-paper-finder?\n`asta-paper-finder` is invaluable for researchers and developers interested in the core mechanics of an LLM-powered paper-finding agent. Its \"frozen-in-time\" nature ensures stability and consistency, making it ideal for:\n- **Reproducibility**: Accurately reproduce and verify evaluation results reported for the Paper Finder agent.\n- **Understanding Core Logic**: Study the pipeline of components, LLM decisions, and relevance judgments without the complexities of a live, evolving product.\n- **Benchmarking**: Use a stable version for comparative analysis and developing new benchmarks for paper-seeking agents.\nThis repository offers a focused view on the single-turn paper search functionality, stripped of multi-turn interaction, UI, and production environment integrations, providing a clean slate for academic and experimental purposes.\n\n## Links\n- **GitHub Repository**: [https://github.com/allenai/asta-paper-finder](https://github.com/allenai/asta-paper-finder){:target=\"_blank\"}\n- **Live PaperFinder Agent**: [https://paperfinder.allen.ai/](https://paperfinder.allen.ai/){:target=\"_blank\"}","metrics":{"detailViews":0,"githubClicks":2},"dates":{"published":null,"modified":"2026-04-24T16:38:25.000Z"}}