Migration Guide

Migrate from Mem0 to AgentBay

Switch in under 10 minutes. This guide maps every Mem0 API call to its AgentBay equivalent.

Step 1: Install

Replace the Mem0 package:

# Remove Mem0
pip uninstall mem0ai

# Install AgentBay
pip install agentbay

Step 2: Initialize

Mem0 (before)

from mem0 import Memory

m = Memory()
# or with API key:
m = Memory.from_config({
    "api_key": "m0-xxx"
})

AgentBay (after)

from agentbay import AgentBay

ab = AgentBay()  # local mode
# or with API key:
ab = AgentBay(
    api_key="ab_live_xxx",
    project_id="my-project"
)

AgentBay with no arguments starts in local mode (SQLite). No signup needed to get started.

Step 3: Store Memories

Mem0

m.add(
    "The user prefers dark mode",
    user_id="user1",
    metadata={"category": "preference"}
)

AgentBay

ab.memory.store(
    title="User prefers dark mode",
    content="The user prefers dark mode",
    type="PATTERN",
    tags=["preference", "ui"]
)

AgentBay separates title from content for better search. The type field (PATTERN, PITFALL, DECISION, etc.) helps organize knowledge.

Step 4: Recall Memories

Mem0

results = m.search(
    "dark mode",
    user_id="user1"
)
for r in results:
    print(r["memory"])

AgentBay

results = ab.memory.recall(
    "dark mode"
)
for entry in results.entries:
    print(f"{entry.title}: {entry.content}")
    print(f"  confidence: {entry.confidence}")

AgentBay recall returns confidence scores (computed at query time with decay), search strategy metadata, and token counts for context-window budgeting.

Step 5: Auto-Memory Chat

Mem0

# Mem0 doesn't have a built-in
# chat wrapper. You must manually:
# 1. Search for relevant memories
# 2. Inject into system prompt
# 3. Call your LLM
# 4. Store the response

AgentBay

reply = ab.brain.chat(
    "How does auth work?",
    provider="openai",
    model="gpt-4o",
    provider_api_key="sk-..."
)
# Auto: recall → inject → LLM → store
print(reply.message)
print(f"Used {len(reply.memories_used)} memories")

Full API Mapping

Mem0	AgentBay	Notes
m.add(text, user_id)	ab.memory.store(title, content, type)	title + content for better search
m.search(query, user_id)	ab.memory.recall(query)	Returns confidence + strategies
m.get_all(user_id)	ab.memory.health()	Health stats, not raw dump
m.delete(memory_id)	ab.memory.forget(entry_id)	Soft delete (recoverable)
m.update(memory_id, data)	ab.memory.store(...)	Auto-dedup by title+type
m.history(memory_id)	Brain time machine	Snapshots + rollback
(no equivalent)	ab.memory.verify(id)	Reset confidence decay
(no equivalent)	ab.memory.compact()	Archive stale, merge dupes
(no equivalent)	ab.brain.chat(msg)	Auto-memory LLM wrapper

What You Gain by Switching

Better search: 4-strategy RRF fusion instead of vector-only
Confidence decay: Old memories fade, frequently-used ones stay strong
Poison detection: 20+ patterns block prompt injection in stored memories
Local mode: Work offline, sync to cloud when ready
brain.chat(): Auto-memory wrapping for any LLM in one line
Memory tiers: Working (24h) → episodic → semantic → procedural (365d)
Multi-agent teams: Shared knowledge with role-based permissions
117 MCP tools: Works with Claude Code, Cursor, any MCP client

Questions about migrating?

Start with local mode to test without commitment.

Quickstart Full Comparison