OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
Agent Memory & RAG

Sr Agents

Sr Agents is an open-source Python research toolkit and benchmark for skill-retrieval-augmented LLM agents. It provides SRA-Bench data, retrieval and skill-use baselines, and a CLI pipeline for retrieving skills, running inference and evaluating task outcomes.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Sr Agents — OpenViST project profile in Agent Memory & RAG.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use Sr Agents when studying whether retrieving reusable capabilities—such as tool procedures, instructions or code—helps an LLM agent complete tasks, or when comparing skill-retrieval approaches on SRA-Bench.

Who it is for

  • Researchers comparing methods for retrieving and applying capabilities in LLM agents.
  • Developers experimenting with skill libraries and retrieval-based agent workflows.

Documented uses

  • Retrieve task-relevant skills from a larger capability library instead of placing every skill in an agent's prompt.
  • Compare retrieval and skill-use methods using benchmark data.
  • Evaluate retrieval rankings and downstream task performance through separate pipeline stages.
Before you choose

Requirements & limits

The public GitHub repository is MIT-licensed. The documented setup requires Python 3.10–3.12; inference uses an OpenAI-compatible chat endpoint.

  • The project focuses on retrieving agent capabilities, not only factual knowledge.
  • SRA-Bench includes six task families and evaluates retrieval, skill incorporation and end-task execution.
  • Inference depends on an OpenAI-compatible endpoint; hosted APIs and local model servers are both described.
  • Running ToolQA requires an additional external corpus.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Run the documented TheoremQA retrieval and evaluation workflow

Use the README's BM25 and Full-Skill Injection example to retrieve a skill, generate an answer through an OpenAI-compatible endpoint, and evaluate it.

Before you start

  • Python 3.10–3.12 and the SR-Agents repository installed with pip install -e . or uv sync.
  • The repository benchmark data, with the skill corpus unzipped under data/bench/corpus/.
  • An OpenAI-compatible chat endpoint and served model identifier; set OPENAI_API_KEY for endpoints that require authentication.

Steps

  1. Retrieve BM25 top-50 skills for the TheoremQA instances using sragents retrieve.

  2. Run sragents infer with the retrieval output as the top-K provider source and k=1, using the endpoint's model identifier and API base.

  3. Run sragents evaluate on the inference JSONL output against the TheoremQA instances.

Expected outcome

Retrieval rankings, per-instance model outputs, and an evaluation JSON reporting end-task accuracy and details.

Guide limitations

  • Inference requires an OpenAI-compatible chat endpoint.
  • The documented example uses TheoremQA, BM25 and Full-Skill Injection; it does not establish results for other configurations.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
13 October 2026
Original website title
oneal2000/SR-Agents · Official homepage title · retained in its source language
How we review sourcesSuggest a correction