Overview

Memorizz helps you build AI agents that can remember, retrieve, and coordinate over time. It combines a configurable memory architecture with provider-backed persistence so agents can run beyond a single prompt window.

Core Building Blocks

src/memorizz/
├── memagent/                # Agent runtime + builder APIs
├── long_term/        # semantic, procedural, episodic memory systems
├── short_term_memory/       # working memory + semantic cache
├── coordination/            # shared memory for multi-agent workflows
├── memory_provider/         # Oracle, MongoDB, filesystem, custom providers
├── internet_access/         # Tavily / Firecrawl / offline web providers
├── sandbox/                 # E2B / Daytona / GraalPy execution providers
└── ui/                      # Local FastAPI-based web UI

Key Capabilities

Capability	What You Get
Persistent agent state	Conversations, summaries, tools, and agent config persisted in a provider
Memory-mode presets	assistant, workflow, and deep_research mode defaults
Semantic retrieval	Embedding-based similarity search for relevant memory recall
Operational tooling	Semantic cache, context-window stats, and auto-summarization support
Extensibility	Custom memory providers and custom internet/sandbox providers

Requirements

• Python 3.7+
• At least one LLM provider (for example OpenAI)
• A persistence backend (filesystem, Oracle, MongoDB, or custom MemoryProvider)

Next Steps

1. Follow Python SDK Quickstart to run your first agent.
2. Use the Local UI Guide if you prefer setting up and operating agents from the browser.
3. Review Concepts for memory and mode mappings.
4. Pick a backend under Memory Providers.