AM-01 is the universal memory protocol for AI agents. Store, retrieve, and share long-term memory across LangChain, CrewAI, AutoGen — without framework lock-in.
from am01_langchain import AM01Memory
memory = AM01Memory(
api_key="am01_your_key",
namespace="support-team",
agent_id="agent-v3"
)
# Done — memory is now persistent across all agent sessions
Every agent framework stores memory differently. Switch from LangChain to CrewAI? Your agent forgets everything. That's not a bug — it's vendor lock-in.
LangChain memories can't talk to CrewAI. Each framework is an island — your agent's context doesn't travel with you.
Switching frameworks wipes accumulated agent knowledge. Years of learned patterns, gone in a deployment.
Healthcare, finance, legal — none of the frameworks have HIPAA, GDPR, or SOC2 memory modes built in.
Multiple agents in a team can't share context. Every agent starts from scratch — even when teammates already know.
AM-01 defines exactly four canonical memory types — enough to capture all agent memory needs, without over-engineering.
Records of specific past interactions and outcomes. What happened, when, and what was the result.
"User alice@co.com asked about
enterprise pricing. Escalated to
sales. Demo scheduled April 20."
Learned facts, concepts, and patterns extracted from interactions. What the agent knows.
"Customer prefers detailed
explanations with code examples
over brief summaries."
Behaviors and strategies that worked. How the agent should act in a given situation.
"When billing query: check tier
first, then provide plan-specific
pricing. Solved 89% of queries."
Agent persona, communication style, and learned preferences. Who the agent is.
"Support agent. Analytical,
helpful. Always provides code
examples. Traits: precise."
{
"$schema": "https://json-schema.org/draft/07",
"title": "AM-01: AI Agent Memory Protocol",
"version": "1.0.0",
"definitions": {
"MemoryType": {
"enum": ["episodic", "semantic",
"procedural", "identity"],
"description": "Canonical AM-01 memory types"
},
"Memory": {
"type": "object",
"required": ["id", "type", "content",
"agent_id", "namespace",
"content_hash", "created_at"],
"properties": {
"id": { "type": "string", "format": "uuid" },
"type": { "$ref": "#/definitions/MemoryType" },
"content": { "type": "string", "maxLength": 65536 },
"metadata": { "$ref": "#/definitions/MemoryMetadata" },
"content_hash":{ "pattern": "^[a-f0-9]{64}$" },
"ttl": { "type": ["integer", "null"] }
}
}
}
}
A complete memory infrastructure layer — not just storage, but intelligence.
Episodic, semantic, procedural, and identity — a complete taxonomy for all agent memory use cases. Schema is open and extensible.
Adapters for LangChain, CrewAI, AutoGen, and LlamaIndex. Memory travels with you — no matter what framework you use.
Find relevant memories instantly. Query by natural language — the API returns ranked results by relevance score.
HIPAA, GDPR, and SOC2 compliant memory storage. PHI isolation, audit logs, right-to-deletion — built in.
See what your agents are learning. Memory growth, type distribution, retrieval latency — all in one dashboard.
AM-01 schema is MIT licensed. The protocol spec is open — anyone can implement it, contribute, or fork.
Make a request right here. No account required to explore the spec.
/v1/memories
/v1/memories/search?q=billing&namespace=demo&limit=5
/v1/memories?namespace=demo&per_page=10
/v1/memories/stats?namespace=demo
Install the adapter for your framework. Your agent's memory is now portable.
Implements BaseMemory interface. Auto-syncs conversation history, semantic search, and identity memory.
pip install am01-langchain
Role-based agent memory. Share memories across agent crews — each agent sees what's relevant to its role.
pip install am01-crewai # TBD
State + event persistence for Microsoft AutoGen agents. Enterprise-grade, with SOC2 compliance mode.
pip install am01-autogen # TBD
RAG-centric semantic memory. AM-01 adds episodic and procedural layers on top of your existing RAG pipeline.
pip install am01-llamaindex # TBD
No credit card required for Sandbox. All plans include the full AM-01 protocol.
Unlimited agents, custom compliance, on-premise deployment, dedicated support, and SLA guarantees.
Get your free API key. No credit card. No framework lock-in. Just universal agent memory.
pip install am01-langchain