Skip to content
Novyx Core Capability

Agent SDK

Build AI agents with built-in persistent memory — auto-recall context before every response, auto-remember after.

bash
pip install novyx-agent

Your agent remembers everything — across sessions, restarts, and deployments. No manual context management.

OpenAIAnthropicLiteLLMFree tier included

Before & After

Without Agent SDK

Manual memory

1
RetrieveQuery memory before every call
2
InjectFormat and stuff into system prompt
3
Call LLMSend to OpenAI/Anthropic
4
ParseExtract what to remember
5
StoreWrite back to memory store
With Agent SDK

Automatic

Auto-recall relevant memories
Inject into system prompt
Call LLM + execute tools
Auto-remember the interaction

One call. Memory handled.

Quick Start

python
from novyx_agent import Agent

agent = Agent(
    api_key="nram_...",
    provider="openai",
    model="gpt-4o",
    system_prompt="You are a helpful assistant with persistent memory.",
    auto_recall=True,
    auto_remember=True,
)

# Agent automatically recalls relevant memories before responding
# and remembers the interaction afterward
response = agent.chat("What did we discuss yesterday about the API redesign?")

That's it. The agent recalls relevant context before every response and stores important information after.

How It Works

Auto-Recall

Before every response, the agent queries Novyx for relevant memories and injects them as tiered context — [strong], [moderate], [background] — so the LLM knows what to trust.

LLM + Tools

The agent calls your chosen LLM (OpenAI, Anthropic, or any LiteLLM model) and executes any @tool functions.

Auto-Remember

After the response, the agent stores the interaction as a memory — available for future recall.

Custom Tools

Give your agent callable functions with the @tool decorator. The agent decides when to call them.

python
from novyx_agent import Agent, tool

@tool(description="Look up a customer by email")
def lookup_customer(email: str) -> dict:
    return db.customers.find(email=email)

agent = Agent(
    api_key="nram_...",
    provider="anthropic",
    model="claude-sonnet-4-20250514",
    tools=[lookup_customer],
)

response = agent.chat("Find the account for alice@example.com")

Architecture

Your Code

agent.chat()

Auto-Recall

Novyx API

LLM + Tools

OpenAI / Anthropic

Auto-Remember

Novyx API

Key Features

Tiered recall — Memories ranked as [strong], [moderate], [background] — background capped at 2
Auto-remember — Stores important context after every response
@tool decorator — Define callable functions with type hints and descriptions
Multi-provider — OpenAI, Anthropic, or any LiteLLM-compatible model
AsyncAgent — Full async support for concurrent agent workflows
Configurable recall — top-k depth, importance scoring, tag filtering
Signal-strength injection — Tiered memories injected into system prompt — LLM sees relevance, not a flat list
All Novyx features — Rollback, audit, knowledge graph, spaces — built in

Async Support

python
from novyx_agent import AsyncAgent

agent = AsyncAgent(
    api_key="nram_...",
    provider="openai",
    model="gpt-4o",
    auto_recall=True,
    auto_remember=True,
)

response = await agent.chat("Summarize yesterday's findings")

Pricing

Works on every tier

The Agent SDK is free. Memory limits are governed by your Novyx API key's tier.

Free: 5,000 memoriesStarter: $12/moPro: $39/mo
See full pricing →

Agents that remember everything.

One import. Persistent memory, auto-recall, auto-remember. Build agents that get smarter over time.

bash
pip install novyx-agent

Start Free

Enter your email to create your developer account.

View on PyPI