mrmemory 0.4.1

Agent Memory Relay — persistent long-term memory for AI agents

MrMemory — Agent Memory Relay

Persistent long-term memory for AI agents. One line to install, three lines to integrate.

Docs · API Reference · Website

Install

pip install mrmemory

# With LangChain/LangGraph support:
pip install mrmemory[langchain]

Quickstart

from amr import AMR

amr = AMR("amr_sk_...")  # or set AMR_API_KEY env var

# Store a memory
amr.remember("User prefers dark mode and vim keybindings")

# Semantic recall
memories = amr.recall("What are the user's preferences?")
for m in memories:
    print(m.content, m.score)

# Forget a memory
amr.forget(memories[0].id)

LLM Auto-Remember

Extract memories from conversations automatically using GPT-4o-mini:

# Extract and store memories from a conversation
result = amr.auto_remember([
    {"role": "user", "content": "I love hiking and my favorite language is Rust"},
    {"role": "assistant", "content": "Great choices!"},
], sync=True)

print(result)  # {"extracted": 2, "created": 2, "duplicates_skipped": 0, ...}

Supports async mode (fire-and-forget), deduplication, and BYOK (bring your own OpenAI key).

Memory Compression

Compress related memories into denser representations:

# Compress memories in a namespace (dry run first)
result = amr.compress(namespace="default", sync=True, dry_run=True)
print(f"Would compress {result['groups_compressed']} groups")

# Actually compress
result = amr.compress(namespace="default", sync=True)
print(f"Reduced {result['before_count']} → {result['after_count']} memories")

Self-Edit Tools

Update, bulk-delete, and merge memories — giving agents full control over their own memory:

# Update an existing memory (re-embeds automatically)
amr.update("mem_abc123", content="Updated: user now prefers light mode", tags=["preference"])

# Bulk delete old memories (dry run first)
result = amr.delete_outdated(older_than_seconds=86400 * 30, dry_run=True)
print(f"Would delete {result['deleted']} memories")

# Actually delete
amr.delete_outdated(older_than_seconds=86400 * 30, tags=["ephemeral"])

# Merge related memories into one (LLM summarization)
merged = amr.merge(["mem_abc123", "mem_def456", "mem_ghi789"])
print(merged.content)  # Summarized by GPT-4o-mini

# Or provide your own merged content
merged = amr.merge(["mem_abc123", "mem_def456"], content="User prefers dark mode and vim keybindings")

Merged memories get is_compressed: true and merged_from tracking in the response.

LangChain / LangGraph Integration

Drop-in checkpointer and store for LangGraph:

from mrmemory.langchain import MrMemoryCheckpointer, MrMemoryStore
from langgraph.graph import StateGraph

checkpointer = MrMemoryCheckpointer(api_key="amr_sk_...")
store = MrMemoryStore(api_key="amr_sk_...")

graph = StateGraph(...).compile(checkpointer=checkpointer, store=store)

Async Support

from amr import AsyncAMR

async with AsyncAMR("amr_sk_...") as amr:
    await amr.remember("User prefers dark mode")
    memories = await amr.recall("What does the user prefer?")

Configuration

amr = AMR(
    api_key="amr_sk_...",       # or set AMR_API_KEY env var
    agent_id="my-assistant",    # default agent ID
    namespace="default",        # default namespace
    timeout=10.0,               # seconds
    max_retries=3,              # retry on transient failures
)

API Endpoints

All requests go to https://amr-memory-api.fly.dev.

Method	Endpoint	Description
POST	/v1/remember	Store a memory
POST	/v1/recall	Semantic search
DELETE	/v1/forget/:id	Delete a memory
GET	/v1/memories	List all memories
POST	/v1/memories/auto	LLM auto-remember from conversations
POST	/v1/memories/compress	Compress related memories
PATCH	/v1/memories/:id	Update a memory
DELETE	/v1/memories/outdated	Bulk delete by age/tags
POST	/v1/memories/merge	Merge memories into one
GET	/v1/ws	WebSocket real-time events

Auth: Authorization: Bearer amr_sk_...

Pricing

Starts at $5/mo — 10K memories, 50K API calls. Sign up →

Links

• Docs: https://mrmemory.dev/docs
• Dashboard: https://mrmemory.dev
• GitHub: https://github.com/masterdarren23/mrmemory

License

MIT