memoryos

[](https://github.com/ninja-otaku/memoryos/actions/workflows/ci.yml) [](https://pypi.org/project/memoryos/) [](LICENSE) [](https://python.org)

Episodic + semantic memory backend for long-running AI agents. Think "hippocampus as a service."

from memoryos import MemoryOS
mem = MemoryOS()
mem.remember("User prefers concise answers.", importance=0.8)
results = mem.recall("How should I answer?")

pip install memoryos
# with vector search (ChromaDB + sentence-transformers):
pip install "memoryos[full]"

from memoryos import MemoryOS

mem = MemoryOS()
mem.remember("The user is debugging a FastAPI timeout issue.", importance=0.7)

for r in mem.recall("FastAPI performance"):
    print(r.content, r.score)

mem.consolidate()  # episodic -> semantic
mem.forget(older_than_days=30, min_importance=0.3)

memoryos-server   # http://localhost:8000

MemoryOS
├── EpisodicMemory   SQLite -- timestamped events, Ebbinghaus decay
├── SemanticMemory   ChromaDB -- consolidated facts, confidence scores
└── ProceduralMemory SQLite -- action templates, success-rate tracking

| Framework | Adapter | |------------|-----------------------| | LangChain | MemoryOSChatMemory | | LangGraph | MemoryOSNode | | CrewAI | MemoryOSTool | | HTTP | FastAPI REST server | | MCP | tools.json |

See CONTRIBUTING.md. PRs welcome!

Apache 2.0 -- see LICENSE.