Open Source · Apache 2.0

The Memory Layer for Multi-Agent Systems

Persistent, semantic memory that survives context window resets. Multi-agent coordination. Self-improving agents. Production-ready.

$ pip install aegis-memory

Get Started View on GitHub

quickstart.py

from aegis_memory import SmartMemory

# Zero-config intelligent memory
memory = SmartMemory(
    aegis_api_key="your-key",
    llm_api_key="your-openai-key"
)

# Automatically extracts what matters
memory.process_turn(
    user_input="I'm a Python dev from Manchester",
    ai_response="Nice to meet you!",
    user_id="user_123"
)

Context dies between sessions

40-80% of production agent failures trace back to memory coordination issues, not model capability.

Context Collapse

LLM rewrites destroy accumulated knowledge. Step 60: 18k tokens. Step 61: 122 tokens. Everything your agent learned — gone.

No Coordination

Agents can't share what they've learned. Duplicate work. Contradictory outputs. Each agent is an island.

Memory Pollution

Without voting, bad memories persist and degrade performance over time. Noise accumulates. Quality decays.

Built for agents that actually work together

Not another vector database. An agent-native memory fabric with research-backed patterns.

Zero Config

Smart Memory

Auto-extracts what matters from conversations. 70% cost savings vs storing everything.

Multi-Agent

Agent Scopes

Private, shared, and global access control. Agents see only what they should.

ACE Pattern

Memory Voting

Track which memories help vs harm task completion. Agents self-improve over time.

< 80ms

Fast

30-80ms queries on 1M+ memories. pgvector HNSW index. Production-grade latency.

Own It

Self-Hostable

Docker, Kubernetes, any cloud. Your data stays yours. Apache 2.0 license.

Metrics

Observable

Prometheus metrics built-in. Structured logging. Know what your memory layer is doing.

What can I build with Aegis Memory?

Discover project ideas powered by persistent agent memory.

Complexity

Audience

Works the way you work

From zero-config auto-extraction to manual fine-grained control. Your choice.

basic.py

from aegis_memory import SmartMemory

# Initialize with zero config
memory = SmartMemory(
    aegis_api_key="your-key",
    llm_api_key="your-openai-key"
)

# Automatically extracts & stores what matters
memory.process_turn(
    user_input="I prefer dark mode and concise answers",
    ai_response="Got it! I'll keep that in mind.",
    user_id="user_123"
)

# Retrieve relevant context for any query
context = memory.get_context(
    "What are the user's preferences?",
    user_id="user_123"
)

crewai.py

from aegis_memory.integrations import AegisCrewMemory
from crewai import Agent, Task, Crew

# Drop-in memory for CrewAI
memory = AegisCrewMemory(
    aegis_api_key="your-key"
)

researcher = Agent(
    role="Researcher",
    goal="Find relevant information",
    memory=memory
)

writer = Agent(
    role="Writer",
    goal="Write compelling content",
    memory=memory  # Shared memory!
)

crew = Crew(
    agents=[researcher, writer],
    tasks=[...],
    memory=memory
)

langchain.py

from aegis_memory.integrations import AegisLangChainMemory
from langchain.chains import ConversationChain
from langchain_openai import ChatOpenAI

# Drop-in replacement for LangChain memory
memory = AegisLangChainMemory(
    aegis_api_key="your-key",
    user_id="user_123"
)

chain = ConversationChain(
    llm=ChatOpenAI(),
    memory=memory  # Persistent across sessions
)

# Memory survives restarts
response = chain.predict(
    input="Remember, I'm working on Project X"
)

manual.py

from aegis_memory import AegisClient

client = AegisClient(api_key="your-key")

# Store with scoped access control
client.add(
    "User prefers dark mode",
    agent_id="ui-agent",
    scope="agent-shared"  # Other agents can see
)

# Semantic search
memories = client.query(
    "user preferences",
    agent_id="ui-agent",
    top_k=5
)

# Vote on memory quality
client.vote(
    memory_id=memories[0].id,
    vote="helpful",
    voter_agent_id="qa-agent"
)

Works with your stack

CrewAI LangChain LangGraph AutoGen

How Aegis compares

Built for multi-agent systems from day one. Not retrofitted.

Feature	Aegis	mem0	Supermemory
Multi-agent scopes	✓	✕	✕
Memory voting (ACE)	✓	✕	✕
Smart extraction	✓	✓	✕
Self-hostable	✓	~	✕
Apache 2.0	✓	✓	✕
Session recovery	✓	✕	✕
Structured handoffs	✓	✕	✕
Framework integrations	✓	✓	✓

"Aegis Memory solved the multi-agent coordination problem we've been struggling with for months."

Ivan Djordjevic — Leidos

Open Source. Self-Hosted. Your Data.

Apache 2.0 licensed. Deploy anywhere. No vendor lock-in. Full data sovereignty.

$ pip install aegis-memory

Get Started Read the Docs Star on GitHub

Apache 2.0 License · v1.2.x