This repository provides Agent KB, a framework for agentic problem-solving that leverages cross-domain experience through a hierarchical memory structure. It's designed for researchers and developers working on autonomous AI agents that require generalization across diverse tasks like QA, coding, and planning.

How It Works

Agent KB employs a hierarchical memory system combining working memory, episodic memory, and a semantic knowledge base. This structure supports autonomous decision-making and planning by LLMs, enabling agents to adapt and generalize across different task domains by drawing on past experiences and structured knowledge.

Quick Start & Requirements

• GAIA: pip install -r requirements.txt, pip install -e ../../.[dev]. Requires GAIA dataset in ./data/gaia/. Set SERP_API_KEY, OPENAI_BASE_URL, and OPENAI_API_KEY environment variables. Run with python run_gaia.py --model-id openai:gpt-4.1 --model-id-search openai:gpt-4.1 --run-name gpt-4.1-gaia.
• SWE-bench: Navigate to ./Agent-KB-SWE-bench/scripts. Use provided shell scripts like run_swe_bench_hints_agentless_repo.sh. Docker environment setup available via build_env.sh.
• Agent KB Service: Format knowledge base in ./agent_kb/agent_kb_database.json. Run service with python ./agent_kb/agent_kb_service.py.

Highlighted Details

• Supports GAIA and SWE-bench benchmarks.
• Modular design for integration with various environments.
• Includes agentless inference scripts with different hint sources.
• Provides baseline inference scripts for comparison.

Maintenance & Community

This project builds upon and adapts code from smolagents and OpenHands. No specific community links (Discord/Slack) or roadmap are provided in the README.

Licensing & Compatibility

The repository does not explicitly state a license. It acknowledges adaptation of code from smolagents and OpenHands, whose licenses should be consulted for compatibility.

Limitations & Caveats

The README does not specify licensing details, which may impact commercial use or closed-source integration. Setup requires specific API keys and dataset preparation, with no explicit mention of supported operating systems or hardware beyond the need for potential GPU/CUDA for LLM inference.