totalrecallagent 0.2.2

On-chain memory infrastructure for AI agents — built on the Internet Computer

total-recall

On-chain memory infrastructure for AI agents — built on the Internet Computer.

Your agent wakes up fresh every session. Total Recall gives it permanent, encrypted, on-chain memory — no cloud, no servers, no single point of failure.

Live at: www.totalrecallagent.com

---

Install

pip install total-recall

With async support:

pip install total-recall[async]

With LangChain integration:

pip install total-recall[langchain]

---

Quick Start

from total_recall import TotalRecallClient

# 1. Create a client (generate your API key at totalrecallagent.com)
memory = TotalRecallClient(api_key="tr_your_key_here")

# 2. Store memory
memory.store("last_context", {
    "user":   "MTR",
    "task":   "HVAC layout review",
    "status": "in_progress",
})

# 3. Retrieve it next session
ctx = memory.get("last_context")
print(ctx.value)      # {"user": "MTR", "task": "HVAC layout review", ...}
print(ctx.updated_at) # datetime object

That's it. Three lines. Your agent remembers.

---

API Reference

TotalRecallClient(api_key, *, base_url, timeout, max_retries)

Param	Type	Default	Description
api_key	str	—	API key from your dashboard
base_url	str	prod	Override API endpoint
timeout	float	30.0	Request timeout in seconds
max_retries	int	3	Retry attempts on network errors

---

memory.store(key, value, tags=[])

Store any value — string, dict/list (auto JSON-encoded), or raw bytes.

memory.store("session_state", {"step": 3, "done": False}, tags=["session"])
memory.store("raw_note", "Agent resumed at checkpoint alpha")

---

memory.get(key)

Retrieve a memory entry. Returns None if not found.

entry = memory.get("session_state")
if entry:
    print(entry.value)      # auto-decoded: {"step": 3, "done": False}
    print(entry.tags)       # ["session"]
    print(entry.updated_at) # datetime

---

memory.get_all()

Get all stored memory entries at once.

entries = memory.get_all()
for e in entries:
    print(e.key, e.value)

---

memory.keys()

List all stored keys.

ks = memory.keys()
# ["session_state", "last_context", "project_notes"]

---

memory.delete(key)

Delete a memory entry. No-op if key doesn't exist.

memory.delete("old_session")

---

memory.merge(key, patch, tags=[])

Merge new data into an existing entry. Creates it if it doesn't exist.

memory.merge("agent_state", {"last_seen": "2026-04-26", "status": "idle"})

---

memory.search(tags)

Search entries by tags. Returns entries that have ALL specified tags.

results = memory.search(tags=["session", "hvac"])
for e in results:
    print(e.key, e.tags)

---

memory.get_stats()

Get current usage stats and tier limits.

stats = memory.get_stats()
print(stats["tier"])                     # "Free" | "Pro" | "Agent" | "Enterprise"
print(stats["storage_bytes"])            # bytes used
print(stats["calls_today"])              # calls today
print(stats["limits"]["calls_per_day"]) # 0 = unlimited

---

memory.ping()

Check if the service is reachable.

status = memory.ping()  # "🧠 Total Recall is alive"

---

Async Usage

from total_recall import TotalRecallAsyncClient

async def run():
    async with TotalRecallAsyncClient(api_key="tr_...") as memory:
        await memory.store("key", {"hello": "world"})
        entry = await memory.get("key")
        print(entry.value)

---

LangChain Integration

Give any LangChain agent persistent on-chain memory:

from total_recall.langchain import TotalRecallMemory
from langchain_openai import ChatOpenAI
from langchain.chains import ConversationChain

memory = TotalRecallMemory(
    api_key="tr_your_key_here",
    session_key="my_agent_session",  # unique per agent/user
)

chain = ConversationChain(
    llm=ChatOpenAI(model="gpt-4o"),
    memory=memory,
    verbose=True,
)

# First session
chain.predict(input="My name is MTR and I work in HVAC.")

# Next session — agent still remembers
chain.predict(input="What do you know about me?")
# → "Your name is MTR and you work in HVAC."

Memory persists across Python processes, machine restarts, and model changes.

---

Real-World Example — AutoGen Agent

import os
import autogen
from total_recall import TotalRecallClient

memory = TotalRecallClient(api_key=os.environ["TOTAL_RECALL_API_KEY"])

def on_agent_start(agent_name: str):
    """Restore agent context at session start."""
    ctx = memory.get_json(f"{agent_name}_context")
    if ctx:
        print(f"[{agent_name}] Resuming. Last task: {ctx.get('last_task')}")
        return ctx
    return {}

def on_agent_end(agent_name: str, state: dict):
    """Persist agent context at session end."""
    memory.merge(f"{agent_name}_context", {
        "last_task":     state.get("current_task"),
        "last_seen":     str(__import__("datetime").datetime.utcnow()),
        "session_count": state.get("session_count", 0) + 1,
    }, tags=["agent", "context"])

---

Real-World Example — OpenAI Assistants

import os
from openai import OpenAI
from total_recall import TotalRecallClient

client = OpenAI()
memory = TotalRecallClient(api_key=os.environ["TOTAL_RECALL_API_KEY"])

# Load memory into system prompt
ctx = memory.get_json("openai_agent_ctx") or {}
system_prompt = f"""You are a helpful assistant.
Previous context: {ctx}
Always update your memory by noting key facts learned each session."""

response = client.chat.completions.create(
    model="gpt-4o",
    messages=[
        {"role": "system",  "content": system_prompt},
        {"role": "user",    "content": "What HVAC projects are we working on?"},
    ]
)

# Save updated context after each session
memory.merge("openai_agent_ctx", {
    "last_response_preview": response.choices[0].message.content[:200],
    "last_seen": str(__import__("datetime").datetime.utcnow()),
})

---

How It Works

• API key is generated on-chain, tied to your Internet Identity
• Memory stored in an ICP canister — no servers, no cloud
• Data persists across upgrades via stable storage
• Agents authenticate with API keys, no Internet Identity needed
• All calls go directly to IC boundary nodes

---

Canister Info

Backend	fwyts-iiaaa-aaaaj-a6lpq-cai
Network	ICP Mainnet
Built with	Motoko, dfx 0.31.0

---

License

MIT — Cleo 3 LLC