The memory engine for multi-agent systems.

Aegis Memory is a production-ready, self-hostable memory engine designed to give AI agents a persistent learning loop. By combining semantic search, scope-aware access control, and ACE (Agentic Context Engineering), Aegis allows agents to share state, vote on strategies, and extract actionable reflections from their failures.

git clone https://github.com/quantifylabs/aegis-memory.git
cd aegis-memory

export OPENAI_API_KEY=sk-...
docker-compose up -d

# Verify
curl http://localhost:8000/health

pip install aegis-memory

# First-time setup
aegis config init

# Check connection
aegis status

# Add your first memory
aegis add "User prefers concise responses" -a assistant -s global

# Query memories
aegis query "user preferences"

# View stats
aegis stats

Aegis provides a powerful CLI for all memory operations:

# Core Operations
aegis add "content"              # Add a memory
aegis query "search text"        # Semantic search
aegis get <id>                   # Get single memory
aegis delete <id>                # Delete memory
aegis vote <id> helpful          # Vote on memory

# ACE Patterns
aegis playbook "error handling"  # Query proven strategies
aegis progress show <session>    # View session progress
aegis features list              # Track feature status

# Data Management
aegis export -o backup.jsonl     # Export memories
aegis import backup.jsonl        # Import memories
aegis stats                      # Namespace statistics

→ Full CLI Reference

from aegis_memory import AegisClient

client = AegisClient(api_key="dev-key", base_url="http://localhost:8000")

# Add a memory
client.add("User prefers concise responses", agent_id="assistant")

# Query memories
memories = client.query("user preferences", agent_id="assistant")

# Vote on usefulness (ACE pattern)
client.vote(memories[0].id, "helpful", voter_agent_id="assistant")

# Cross-agent memory sharing
client.add(
    content="Task: Build login. Steps: 1) Form, 2) Validation, 3) API",
    agent_id="planner",
    scope="agent-shared",
    shared_with_agents=["executor"]
)

Aegis Memory is not just another vector database. It's an active strategy engine with primitives designed to turn agent execution into persistent organizational intelligence.

• Semantic Search — pgvector HNSW index for O(log n) queries at scale
• Scope-Aware Access — agent-private, agent-shared, global with automatic ACL
• Multi-Agent Handoffs — Structured state transfer between agents
• Auto-Deduplication — Hash-based O(1) duplicate detection

• Memory Voting — Track which memories help vs harm task completion
• Delta Updates — Incremental changes that prevent context collapse
• Reflections — Store insights from failures for future reference
• Session Progress — Track work across context windows
• Feature Tracking — Prevent premature task completion

• Self-Hostable — Docker, Kubernetes, any cloud
• Observable — Prometheus metrics, structured logging
• Fast — 30-80ms queries on 1M+ memories
• Safe — Data export, migrations, no vendor lock-in

Drop-in support for popular agent frameworks:

# LangChain
from aegis_memory.integrations.langchain import AegisMemory
chain = ConversationChain(llm=llm, memory=AegisMemory(agent_id="assistant"))

# CrewAI
from aegis_memory.integrations.crewai import AegisCrewMemory
crew = Crew(agents=[...], memory=AegisCrewMemory())

→ Integration Guides

Aegis implements patterns from recent research on self-improving agents:

# After a memory helped complete a task
aegis vote <memory-id> helpful -c "Successfully paginated API"

# Query only effective strategies
aegis playbook "API pagination" -e 0.3

# Track work across context windows
aegis progress create build-dashboard -a coder
aegis progress update build-dashboard -c auth -i routing
aegis progress show build-dashboard

→ ACE Patterns Guide

• Quickstart — Get running in 15 minutes
• ACE Patterns — Self-improving agent patterns
• Operations — Backup, monitoring, upgrades
• Design — Technical deep-dive
• Recipes — 10 production-ready patterns
• API Reference — OpenAPI docs (when running)

docker-compose up -d

kubectl apply -f k8s/

→ Full Configuration

We welcome contributions! See CONTRIBUTING.md for guidelines.

# Run tests
pytest tests/ -v

# Run linting
ruff check server/

Apache 2.0 — Use it however you want. See LICENSE.

• GitHub Discussions
• Issue Tracker
• Changelog

Built with ❤️ for the agent community