Aether vs Letta
Letta and Aether solve “my agent forgets between sessions” from opposite directions. Letta is an agent runtime; Aether is the memory layer that sits under the runtime you already use.
01The short version
Letta is an agent runtime, descended from the MemGPT research project, that manages its own memory hierarchy. If you want a full agent operating system and you're happy building inside it, Letta is compelling.
Aether is the memory layer that sits under whatever runtime you already use — LangGraph, the Vercel AI SDK, Semantic Kernel, or your own loop. You keep your framework; Aether keeps the memories, with the database included, so it's one bill, not three.
02Where Letta is genuinely strong
Letta grew out of MemGPT, the research project that treated an LLM's context window like an operating system treats RAM — paging memories in and out of a limited window. That lineage shows, in serious agent-runtime depth: agents as long-running processes, with the system deciding what stays in context and what gets paged out.
If your mental model is “I want a platform that runs my agents and handles their memory as one integrated system,” Letta is one of the most thoughtful implementations of that idea.
03Side by side
| Aether | Letta | |
|---|---|---|
| What it is | Memory layer for agents, on Aether's own storage + retrieval engine | Agent operating system with built-in memory management |
| Where your agent runs | Wherever it already runs — LangGraph, Vercel AI SDK, Semantic Kernel, your own loop | Inside Letta's runtime, for full value |
| Backing store | Included — no separate store to bring or rent | A layer in Letta's stack |
| Language SDKs | Python, TypeScript, Go, and .NET at full parity | No first-party .NET; doesn't cover all four languages |
| Pricing model | Flat monthly rates, no per-API-call metering | Platform pricing — see Letta |
04Memory across sessions, in code
The test that matters: your agent learns something today and recalls it next week, in a different session, without you re-feeding context. Scope memory to a stable id; recall() returns the most relevant memories ranked by score.
from aether import Memory
# Scope memory to a stable user or agent id.
memory = Memory("user-42")
# Today: your agent learns something.
memory.remember("Freezes production deploys after 2pm on Fridays")
# Next week, a fresh session — it remembers, ranked by relevance.
for hit in memory.recall("when can I schedule a deploy?"):
print(hit.score, hit.text)The same surface ships in Python, TypeScript, Go, and .NET at full parity. See the quickstarts for your language.
05Which one fits
When to choose Letta
- You want one platform to run your agents, not just remember for them, and you're comfortable adopting its execution model.
- The MemGPT-style memory hierarchy — the system paging context in and out for you — matches how you think about agents.
- You're starting fresh, with no existing framework investment to protect.
When to choose Aether
- You already have a runtime or framework you like, and you need memory under it — not a new home for your agents.
- You want one bill: memory, document storage, semantic search, and embeddings on one platform.
- You're a .NET or Go shop and first-party SDK parity is non-negotiable.
- You want flat, predictable pricing rather than metered API calls.
06Frequently asked questions
Is Letta a bad choice?
No. If you want an agent operating system with research-grade memory management built in, Letta is a strong one. This page exists because the two products are different shapes, not because one is wrong.
Does Aether replace my agent framework?
No — that's the point. Aether sits under LangGraph, the Vercel AI SDK, Semantic Kernel, or your own loop. Your runtime stays your runtime.
Do I need to bring my own backing store?
No. Aether runs on its own Rust storage and retrieval engine. Storage, semantic search, and embeddings are included — one platform, one bill.
What does it cost?
Flat monthly rates with a free tier and no per-API-call metering — your bill doesn't grow because your agent got chatty. See the pricing page for current tiers.
Your agents need memory. Not three bills.
First-party SDKs for four languages, sub-millisecond recall, one flat rate. Start free — no credit card required.