CAMEL-AI Memory

How it works

1. Install camel-memorysync from PyPI.
2. Pass memory=MemorySyncMemory(...) to your ChatAgent.
3. Every user turn is sent to MemorySync, which extracts the durable facts in it and stores only those; assistant replies are not sent, and your verbatim history stays local.
4. On each step, the user’s relevant memories are recalled and injected after the system prompt — days later, in new sessions, in other frameworks.

Before you start

1. A MemorySync API key. Create one in the dashboard under Settings → API Keys, inside a project.
2. Python 3.10+ with camel-ai 0.2.60 or newer.

Install

One package:

pip install camel-memorysync

Quickstart

Step 1 — hand the agent its memory

from camel.agents import ChatAgent
from camel_memorysync import MemorySyncMemory

memory = MemorySyncMemory(
    user_id="customer-42",      # required — who these memories belong to
    session_id="support",       # session scope — tags this conversation's facts
)
agent = ChatAgent(
    system_message="You are a helpful travel assistant.",
    memory=memory,
)

Step 2 — teach it, then ask later

agent.step("I always prefer window seats on long flights")

# a new session, days later:
agent.step("which seat should I book for the Oslo flight?")  # remembers

How recall enters the context

CAMEL’s ScoreBasedContextCreator sorts records by timestamp. The recall block is ONE SYSTEM-role record with timestamp=0.0, so it lands right after the system prompt — your history is never reordered, rewritten, or role-swapped. An anti-echo filter drops any match already present verbatim in the live context, so the question the user just asked can never come back as a “memory”.

[system]  You are a helpful travel assistant.
[system]  Relevant long-term memories about this user:
          - Prefers window seats on long flights
[user]    which seat should I book for the Oslo flight?

Advanced: the storage seam

If you compose CAMEL memories yourself, MemorySyncStorage slots into CAMEL’s own ChatHistoryMemory as a BaseKeyValueStorage. save() keeps verbatim records locally (exact round-trip) and sends each user turn to MemorySync, which extracts the durable facts in it and stores only those; assistant replies are not sent, and system prompts and tool chatter never leave the process. A retried save is recognised server-side, so each turn is extracted once:

from camel.memories import ChatHistoryMemory, ScoreBasedContextCreator
from camel.types import ModelType
from camel.utils import OpenAITokenCounter
from camel_memorysync import MemorySyncStorage

memory = ChatHistoryMemory(
    context_creator=ScoreBasedContextCreator(
        OpenAITokenCounter(ModelType.GPT_4O_MINI), 2048),
    storage=MemorySyncStorage(user_id="customer-42", session_id="support"),
)

Deletion — designed against data loss

Configuration

Quotas and plan limits

Hitting a monthly plan limit never breaks a conversation. Over-limit writes are accepted without storing and recalls return empty — the agent keeps answering from live history. Evaluation keys instead surface a truthful 429, so limits show up in testing, not production.

Troubleshooting

• Import crash mentioning `FastMCP` — the installed mcp release is newer than camel-ai supports. Reinstall camel-memorysync to restore a compatible version.
• A recall was skipped — the 1.2s budget fails open to history-only, and topics under min_query_chars skip recall by design.
• A “memory” repeats the question just asked — cannot happen: the anti-echo filter drops matches already present in the live context. If you see it, the text differed — check the stored fact.
• `clear()` didn’t delete remote memories — by design. Use forget_session() or forget_user() for explicit, scoped remote deletes.

Supported versions

How it compares

CAMEL ships a Mem0 storage inside its own repo — the only other memory SaaS in the ecosystem. Verified against its source:

Where to go next