Skip to content

Comparison

hippo vs Letta: a Letta alternative for coding agents

Letta, formerly MemGPT, now builds Letta Code, a stateful agent harness whose memory lives in blocks the agent edits itself. Hippo is memory for the agents you already run, such as Claude Code, Codex and Cursor, in one store they share.

Both keep memory where you can see it: Letta Code tracks its context in git, and hippo writes markdown mirrors you can commit. They differ in who edits memory. In Letta, the agent rewrites its own blocks with built-in tools. In hippo, memories come from hooks, the CLI and MCP tools, and you can mark a recalled one wrong or supersede an old fact.

Feature by feature

Every row of the README comparison table, hippo against Letta. The Letta column was last checked against Letta's own pages on 2026-09-28.

hippo and Letta, row by row from the README comparison table
Feature hippo Letta
Decay by default Yes No
Retrieval strengthening Yes No
Reward-proportional decay Yes No
Hybrid search (BM25 + embeddings) Yes ?
Schema acceleration / knowledge graph Yes (schema) No
Conflict detection + resolution Yes No
Multi-agent shared memory Yes Yes (shared memory blocks)
Transfer scoring Yes No
Outcome tracking Yes No
Confidence tiers Yes No
Spatial organization No No
Lossless compression No No
Cross-tool import (ChatGPT/Claude/Cursor) Yes No
Auto-hook install Yes No
MCP server Yes Yes (hosted, needs an API key)
Zero runtime deps Yes No (npm deps)
LongMemEval (best published) 98.0% any / 88.5% all R@5* (local MiniLM; 99.8% any-evidence with voyage-3-large; s_cleaned, per-haystack) N/A
Git-friendly Yes Yes (memory tracked in git)
Framework agnostic Yes Yes
License MIT Apache-2.0

* Hippo's figures are on longmemeval_s_cleaned with a per-question haystack, each the best of five retrieval settings in the benchmark scripts, not hippo recall. Any-evidence R@5 counts a hit when any answer session is in the top 5, over all 500 questions: 98.0% with the free local MiniLM embedder (an optional install) and 99.8% with voyage-3-large (measured 2026-06-09, not re-run). All-evidence R@5 counts a hit only when every answer session is in the top 5, over the 470 questions that have an answer: 86.8 to 88.5% with MiniLM. gbrain first published 97.6%, an any-evidence score over all 500; its report (opens in new tab) now leads with all-evidence, 95.53% (449 of 470) with the paid Voyage rerank-2.5 reranker and 93.19% without it. On all-evidence recall gbrain is ahead. The June 2026 build scored 98.6 any-evidence; docs/evals/2026-09-23-longmemeval-reproduction.md (opens in new tab) has both runs. An older hippo number, 86.8% R@5 on longmemeval_oracle under pooled (non-per-haystack) retrieval, is not comparable to per-haystack figures.

What Letta ships today (checked 2026-09-28)

When to pick which

Pick hippo

You already use Claude Code, Codex or Cursor and want them to share one store of lessons that you can correct, on your machine with no account.

Pick Letta

You want a coding agent whose memory is part of the agent, one that edits its own memory blocks, and you are ready to switch agents to get it.

Get started with hippo All questions

FAQ

hippo and Letta, asked directly

Is Letta the same as MemGPT?

Letta grew out of MemGPT; its GitHub README calls it "Letta (f.k.a. MemGPT)". Active development has moved to Letta Code, a stateful agent harness, and the original Letta server now sits on an archive branch (checked 2026-09-28).

Is hippo a Letta alternative?

It depends on what you want from Letta. Letta Code is a stateful agent harness with its memory built in. Hippo is memory for the agents you already use: Claude Code, Codex, Cursor and any MCP client share one store, on your machine with no account, and you can mark a lesson wrong or supersede an old fact. If you want an agent that manages its own memory blocks, that is Letta Code.