Preview. These maps are our reading of public sources and haven't been checked with the projects' maintainers yet.
Letta (formerly MemGPT)
Originated MemGPT self-editing memory. Now Letta Code: a stateful agent harness whose agents rewrite their own memory blocks, skills and prompts; memory tracked in git (MemFS), with Letta Cloud as default backend.
www.letta.com/Changed direction
letta-ai/letta README now says the V1 API server is retired (archive branch) and current source is letta-ai/letta-code (v0.34.24 released 2026-10-25). Last letta server release 0.16.8 on 2026-05-14. AgentFile (.af) export/import deprecated and removed. (checked 2026-10-25)
In Agent memory
Where it stands apart
- Who is it for? both (14% of 21 others)
- In the default setup, who mainly decides what gets remembered? the agent, by calling memory tools (19% of 21 others)
- What is the main storage model for long-term memory? plain files (Markdown, text, JSON) (15% of 20 others)
- What shape does it mainly take (the one its docs lead with)? agent runtime or harness with memory built in (0% of 21 others)
What it shares with most of the community
- Is it actively maintained? active (67% of 21 others)
- How is the main thing licensed? permissive open source (MIT/Apache/BSD) (67% of 21 others)
- How is it mainly sustained? company with a paid product, tier, or ads (75% of 20 others)
- Can it run entirely on your own device or infrastructure, without the maker's servers? yes (67% of 21 others)
- By default, how do old or wrong memories mainly go away? the system updates, merges, or decays them automatically (58% of 19 others)
- Does it work with models from different LLM vendors? yes, model-agnostic (86% of 21 others)
- Does it publish scores on public memory benchmarks (LoCoMo, LongMemEval, BEAM, etc.)? yes (62% of 21 others)
Nearest neighbors
- Basic Memory 59% of 17 shared answers match
- MemOS (MemTensor) 59% of 17 shared answers match
- Supermemory 59% of 17 shared answers match
Every answer
| Question | Answer | Source |
|---|---|---|
| Is it actively maintained? | active Pivoted from a memory/agent server to an agent harness, but still memory-centric and very active. | source 2026-10-25 |
| How is the main thing licensed? | permissive open source (MIT/Apache/BSD) Apache-2.0 | source 2026-10-25 |
| How is it mainly sustained? | company with a paid product, tier, or ads | source 2026-10-25 |
| Can it run entirely on your own device or infrastructure, without the maker's servers? | yes 'letta backend local' / 'letta server' run locally; Letta Cloud is the default. | source 2026-10-25 |
| Do you need an account to use it? | optional account Can sign in with Letta or proceed locally with own LLM keys; some features need Letta login. | source 2026-10-26 |
| Who is it for? | stands apart both Developers (SDK, CLI) and end users via chat.letta.com, Slack, Telegram. | source 2026-10-25 |
| Which shared standard or format does it mainly build on? | none of these (its own format or interface) Own MemFS tracked in git; Agent File format deprecated. | source 2026-10-26 |
| In the default setup, who mainly decides what gets remembered? | stands apart the agent, by calling memory tools Agents rewrite their own memory; periodic 'dreaming' (sleep-time) also consolidates. | source 2026-10-25 |
| How can the person the memories are about see and edit them? | directly, as readable files or in a UI they control /palace views memory; MemFS keeps context in git, syncable to your own GitHub repo. | source 2026-10-25 |
| By default, how do old or wrong memories mainly go away? | the system updates, merges, or decays them automatically Agent self-edits plus background dreaming/reflection rewrite memory. | source 2026-10-25 |
| Where do memories live in the default setup the docs lead with? | in the vendor's cloud 'Letta Cloud is the default' on first launch. | source 2026-10-25 |
| Does it work with models from different LLM vendors? | yes, model-agnostic | source 2026-10-25 |
| What is the main storage model for long-term memory? | stands apart plain files (Markdown, text, JSON) MemFS: all context, including memory blocks, tracked via git. | source 2026-10-25 |
| What shape does it mainly take (the one its docs lead with)? | stands apart agent runtime or harness with memory built in | source 2026-10-25 |
| Is there a first-party MCP server for it? | no MCP client support, no first-party memory MCP server found. | source 2026-10-25 |
| How is memory separated between users or tenants? | one store per user or agent; the app partitions it Memory is per agent; multi-user handling is up to the deployment. | source 2026-10-25 |
| Does it publish scores on public memory benchmarks (LoCoMo, LongMemEval, BEAM, etc.)? | yes Published LoCoMo results in 'Benchmarking AI Agent Memory: Is a Filesystem All You Need?' | source 2026-10-25 |
Sources
This page was derived from:
- GitHub · github.com/letta-ai/letta-code · read 2026-10-25
- Web page · docs.letta.com/letta-code/memfs · read 2026-10-25
- Web page · www.letta.com/ · read 2026-10-25
- Web page · www.letta.com/blog/benchmarking-ai-agent-memory · read 2026-10-25
Are you a maintainer? Claiming and correcting projects is coming soon.