agentbay 0.4.0

Persistent memory for AI agents. Store, recall, and share knowledge across sessions.

AgentBay Python SDK

Persistent memory for AI agents. 3 lines to give your agent a brain.

Install

pip install agentbay

Quick Start -- Auto-Memory (Recommended)

The chat() method wraps your LLM call with automatic memory. No manual store/recall needed.

from agentbay import AgentBay

brain = AgentBay("ab_live_your_key", project_id="your-project-id")

# Memory happens automatically -- no manual store/recall needed
response = brain.chat([
    {"role": "user", "content": "fix the auth session expiry bug"}
])

# brain.chat() automatically:
# 1. Recalled relevant memories about auth and sessions
# 2. Injected them into the LLM context
# 3. Got the response from Claude
# 4. Extracted learnings and stored them for next time

Using OpenAI

response = brain.chat(
    [{"role": "user", "content": "refactor the payment module"}],
    model="gpt-4o",
    provider="openai",
)

Passing extra LLM parameters

response = brain.chat(
    [{"role": "user", "content": "optimize the database queries"}],
    max_tokens=8192,
    temperature=0.7,
)

Disabling auto-memory

# Recall only (don't store new learnings)
response = brain.chat(messages, auto_store=False)

# Store only (don't inject recalled memories)
response = brain.chat(messages, auto_recall=False)

# No memory at all (just use as a plain LLM wrapper)
response = brain.chat(messages, auto_recall=False, auto_store=False)

Mem0-Compatible API

If you're migrating from Mem0, AgentBay supports the same add() / search() interface:

brain = AgentBay("ab_live_your_key", project_id="your-project-id")

# Store with automatic type detection
brain.add("The auth bug was caused by expired JWT tokens not being refreshed")
brain.add("We decided to use PostgreSQL instead of MongoDB for ACID compliance")

# Search
results = brain.search("authentication issues")
for r in results:
    print(r["title"], r["confidence"])

Manual Memory Control

For full control, use store() and recall() directly:

from agentbay import AgentBay

brain = AgentBay("ab_live_your_key", project_id="your-project-id")
brain.store("Next.js 16 + Prisma + PostgreSQL", title="Project stack")
results = brain.recall("What stack does this project use?")

Or create a new brain on the fly:

from agentbay import AgentBay

brain = AgentBay("ab_live_your_key")
brain.setup_brain("My Agent's Memory")
brain.store("Always use UTC timestamps", title="Convention", type="PREFERENCE")

Core API

Method	What it does
brain.chat(messages, model, provider, ...)	LLM call with automatic memory
brain.add(data)	Store with auto-detection (Mem0-compatible)
brain.search(query)	Search memories (Mem0-compatible alias)
brain.store(content, title, type, tier, tags)	Save a memory (full control)
brain.recall(query, limit, tier, tags)	Search memories (semantic + keyword)
brain.forget(knowledge_id)	Archive a memory
brain.verify(knowledge_id)	Confirm a memory is still accurate
brain.health()	Get memory stats
brain.setup_brain(name, description)	Create a new Knowledge Brain

Memory Types

• PATTERN -- Learned behaviors and recurring themes
• FACT -- Verified information
• PREFERENCE -- User/agent preferences
• PROCEDURE -- Step-by-step processes
• CONTEXT -- Situational context
• PITFALL -- Bugs, errors, and fixes to avoid
• DECISION -- Architecture and design decisions

With CrewAI

pip install agentbay[crewai]

from crewai import Agent
from agentbay.integrations.crewai import AgentBayCrewAIMemory

memory = AgentBayCrewAIMemory(
    api_key="ab_live_your_key",
    project_id="your-project-id",
)

agent = Agent(
    role="Researcher",
    goal="Find and remember information",
    memory=memory,
)

With LangChain

pip install agentbay[langchain]

from langchain_openai import ChatOpenAI
from langchain.agents import initialize_agent, AgentType
from agentbay.integrations.langchain import AgentBayMemoryTool

tool = AgentBayMemoryTool(
    api_key="ab_live_your_key",
    project_id="your-project-id",
)

llm = ChatOpenAI()
agent = initialize_agent(
    tools=[tool],
    llm=llm,
    agent=AgentType.OPENAI_FUNCTIONS,
)
agent.run("Remember that deploys happen every Tuesday at 2pm UTC")

Error Handling

from agentbay import AgentBayError, AuthenticationError, RateLimitError

try:
    results = brain.recall("query")
except AuthenticationError:
    print("Bad API key")
except RateLimitError:
    print("Slow down")
except AgentBayError as e:
    print(f"Error {e.status_code}: {e}")

Environment Variables

For chat(), set your LLM provider API key:

# For Anthropic (default provider)
export ANTHROPIC_API_KEY=sk-ant-...

# For OpenAI
export OPENAI_API_KEY=sk-...

Or pass it directly:

response = brain.chat(messages, api_key="sk-ant-...")

Links

• AgentBay -- AI agent marketplace + memory platform
• API Docs
• MCP Server