reflective-agent-architecture
by angrysky56
Overview
A research prototype for a Reflective Agent Architecture (RAA) that integrates modern associative memory with metacognitive monitoring for insight-like problem-solving, capable of self-reflection and dynamic adaptation.
Installation
python src/server.pyEnvironment Variables
- EMBEDDING_PROVIDER
- EMBEDDING_MODEL
- OLLAMA_BASE_URL
- LMSTUDIO_BASE_URL
- LMSTUDIO_API_KEY
- OPENROUTER_BASE_URL
- OPENROUTER_API_KEY
- OPENROUTER_SITE_URL
- OPENROUTER_APP_NAME
- LLM_PROVIDER
- LLM_MODEL
- OPENAI_API_KEY
- OPENAI_BASE_URL
- ANTHROPIC_API_KEY
- GEMINI_API_KEY
- HF_TOKEN
- NEO4J_URI
- NEO4J_USER
- NEO4J_PASSWORD
- COMPASS_MODEL
- COMPASS_PROVIDER
- MCP_CONFIG_PATH
- LOG_LEVEL
Security Notes
Direct `subprocess.run` calls (e.g., in `_search_codebase`, `_run_mace4`) can be vulnerable to command injection if inputs are not thoroughly sanitized. The `exec` call within `src/compass/sandbox.py` is intended for sandboxing but represents an inherent risk point. The use of `torch.load(weights_only=False)` for loading model projections in `src/vectordb_migrate/migration.py` presents a deserialization vulnerability if untrusted projection files are inadvertently loaded. API keys are generally handled via environment variables, which is good practice.
Similar Servers
inspector
An interactive web UI for inspecting and debugging Model Context Protocol (MCP) servers, allowing users to explore resources, prompts, and tools, manage connections, and facilitate OAuth 2.0 authentication flows.
mcp_massive
An AI agent orchestration server, likely interacting with LLMs and managing multi-agent workflows.
sample-agentic-ai-demos
Provides various examples for building Agentic AI with AWS Bedrock using the Model Context Protocol (MCP) for tool invocation and inter-agent communication.
AgentUp
A developer-first framework for building, deploying, and managing AI agents, bringing Docker-like consistency and operational ease to AI agent development.