Novyx vs Letta
Novyx is memory-as-a-service. Letta (formerly MemGPT) is a full agent framework. Here's how they compare.
TL;DR
Novyx is a composable memory API that plugs into any framework — LangChain, CrewAI, LlamaIndex, MCP, or your own code. You get rollback, Replay, Cortex, knowledge graphs, cross-tenant threat intelligence, auto defense, and a 107-tool MCP server. Every Novyx customer makes every other customer safer. Letta (formerly MemGPT) is an opinionated agent framework with built-in self-managing memory. You use their agent architecture or nothing.
| Feature | Novyx | Letta |
|---|---|---|
| Approach | Memory-as-a-service (composable) | Full agent framework |
| Persistent memory | ✓ | ✓ (via MemGPT architecture) |
| Semantic search | ✓ | ✓ |
| Point-in-time rollback | ✓ (Magic Rollback) | — |
| Cryptographic audit trail | ✓ (SHA-256 + RSA) | — |
| Replay (time-travel debugging) | ✓ | — |
| Cortex (autonomous maintenance) | ✓ | — |
| Self-managing memory | Via Cortex | Via MemGPT architecture |
| Knowledge graph | ✓ | — |
| MCP server | 107 tools | — |
| Threat intelligence | ✓ (cross-tenant) | — |
| Auto defense | ✓ (self-tuning) | — |
| Attack correlation | ✓ | — |
| Framework integrations | LangChain, CrewAI, LlamaIndex, MCP | Own framework |
| Python SDK | ✓ | ✓ |
| JS/TS SDK | ✓ | — |
| Free tier | 5,000 memories | Self-hosted free |
| Eval & CI/CD gate | ✓ | — |
| Self-hosted | Roadmap Q2 2026 | ✓ |
| Open source | — | ✓ |
The Rollback Differentiator
Letta's MemGPT architecture gives agents the ability to manage their own memory — deciding what to remember, what to forget, and how to organize context. But this self-management has no safety net. When an agent makes a bad memory decision — storing a hallucination, overwriting a correct fact, or forgetting something important — there's no built-in way to undo it.
Novyx's Magic Rollback provides that safety net. Every memory operation is recorded in a SHA-256 hash chain, and you can revert to any point in time with a single API call. The cryptographic audit trail proves the rollback was complete and verifiable. This is especially important for self-managing memory systems: the more autonomy you give your agent over its own memory, the more critical it is to have an undo button.
In practice, this means you can give your Novyx-powered agent full autonomy to manage its memory knowing that any mistake is reversible. With Letta, self-managed memory errors compound over time because there's no mechanism to roll back the agent's own memory decisions.
Composable Service vs Monolithic Framework
The fundamental architectural difference: Novyx is a memory API that plugs into any stack. Letta is a complete agent framework that includes memory as one component. If you're already using LangChain, CrewAI, LlamaIndex, or custom agent code, Novyx drops in as a memory backend with dedicated integrations for each framework. Letta requires you to adopt its entire agent architecture.
This matters for teams with existing infrastructure. Adding Novyx to a LangChain agent is three lines of code. Migrating to Letta means rewriting your agent from scratch using Letta's framework, its tool definitions, its message format, and its execution model. For new projects starting from zero, Letta's all-in-one approach can be convenient. For teams with existing agents, Novyx's composability avoids a costly rewrite.
Memory Maintenance: Cortex vs Self-Managing
Both platforms address the problem of memory quality over time, but with different approaches. Letta's MemGPT architecture lets the agent itself decide when to archive, retrieve, and reorganize memories. This is elegant but unpredictable — the agent's memory management is only as good as its current reasoning.
Novyx's Cortex system takes a different approach: autonomous background maintenance cycles that consolidate related memories, reinforce frequently-accessed ones, decay stale information, and surface insights. Cortex operates independently from your agent's reasoning, so memory quality doesn't depend on the agent remembering to manage its own memory. Combined with Replay's time-travel debugging, you can observe exactly how Cortex is maintaining your memory store and verify it's working as expected.
When to Choose Novyx
- ✓You already use LangChain, CrewAI, LlamaIndex, or another framework
- ✓You want memory as a composable service, not a full framework
- ✓You need rollback, Replay, and Cortex for memory safety
- ✓You want a JS/TS SDK alongside Python
- ✓You need MCP integration for Cursor or Claude Code
When to Choose Letta
- ✓You want a complete agent framework with built-in memory management
- ✓You prefer the MemGPT self-managing memory architecture
- ✓You need to self-host and want an open-source solution
- ✓You don't already have a framework and want an all-in-one solution
Code Comparison
from novyx import Novyx
nx = Novyx(api_key="YOUR_API_KEY")
# Works with any framework
nx.remember("User prefers dark mode", tags=["prefs"])
results = nx.recall("user preferences")
# Rollback (Novyx-only)
nx.rollback(target="2026-01-15T10:00:00Z")
# Use with LangChain, CrewAI, LlamaIndex, etc.
from novyx_langchain import NovyxMemory
memory = NovyxMemory(api_key="YOUR_API_KEY")Start Building with Persistent Memory
5,000 memories free. No credit card required.