Co-founder at Basic Memory. Building a knowledge base your AI actually remembers. Open source, MCP-native. Practitioner, not a guru. Texan.
Austin, TX
Joined January 2011
- Tweets103
- Following327
- Followers136
- Likes2.5K
Drew Cain retweeted
Recommended read, and basically a map of what we’ve been building. Basic Memory is agent memory as plain Markdown you own, and last week we shipped the agents that maintain it.
A few things we’d add 👇
Memory is what allows Claude to learn from experience. In a new Engineering Blog post, we share how “dreaming” helps Claude organize what it’s learned, correct its mistakes, and find what it needs more easily the next time. anthropic.com/research/compo…
Drew Cain retweeted
basicmemory.com/blog/context…
Some context about getting to our 1.0 release.
Basic Memory Cloud is 1 year old. (Holy smokes!)
BM is leaving beta. (1.0 baby!!)
We're introducing usage-based pricing. (No change for
most of you, and a real path for heavy users.)
We're introducing Memory Agents. (Knowledge base management made easy!)
Paul is finally full-time at Basic Memory. (Somehow he is even busier.)
Drew Cain retweeted
Here's a video about our Basic Memory Agents. We took a different approach to managed agents.
Agents are just notes in your knowledge base with `type: agent`.
You can define when and how they run.
Agents can run automatically or via your harness with MCP.
All changes are versioned and reviewable. See a diff of changes and accept or rollback.
Agents use DeepSeek v4.1 so they are very inexpensive to run.
You can see detailed usage and control costs.
This is our answer to "dreaming" for AI memory. You control how your memory curates itself, and it just happens. No magic, no surprises.
Drew Cain retweeted
This is a great talk IMO. Its super cool to see Anthropic talking about things we are working on in Basic Memory :).
@lamismukta @ainativedev
It’s only a kludge if you’re doing it that way. Basic Memory handles it for you seamlessly. Obviously LLMs struggle with recall, but I think long term the human-to-AI memory bridge is the real problem to solve.
The standard approach to agent memory is a kludge. Two separate systems: short-term context window management, long-term vector store retrieval. Different controllers, different heuristics, each optimized independently.
A new paper, AgeMem (arxiv 2601.01885), argues memory should be part of the learned policy, not two bolt-on subsystems. The mechanism: expose memory operations as tools. ADD, UPDATE, DELETE for long-term storage. RETRIEVE, SUMMARY, FILTER for short-term context. The agent calls these tools during reasoning loops. Then train the whole policy end-to-end with RL.
The training approach is 3-stage progressive RL using step-wise GRPO, a modified group relative policy optimization that propagates terminal task rewards backward to early memory decisions. Early "store this" or "forget this" actions get credit from final task outcomes, not from intermediate signals.
Results on Qwen2.5-7B: 41.96% average across ALFWorld, SciWorld, PDDL, BabyAI, HotpotQA. Versus Mem0's 37.14%. On Qwen3-4B: 54.31% versus 44.70%.
The products I keep watching in this space: Mem0 (open source), MemoryOS (layered memory with LRU-style eviction), Zep (episodic memory with graph retrieval), A-MEM from Rutgers and Ant Group. What AgeMem adds that none of these do: the agent learns when to forget, not just when to store.
My read: the trajectory leads to agents that manage their own cognitive overhead without prompt engineering.
xitter.cf/beamnxw/status/2086457…
1
38
- "AI-ify all your data via an agent"
- Add it to your long term memory (editable, searchable, prunable)
- Standardized so it is understandable by any other system
Google v0.2 OKF spec looks pretty cool. I'm going to make Basic Memory "OKF native" or whatever that term is so your AI memory is completely portable and interchangeable between agents.
cloud.google.com/blog/produc…
1
12
Drew Cain retweeted
Basic Memory Cloud is finally in both ChatGPT’s Plugin directory and Claude’s Connectors Directory.
Getting through both publishing processes was a quite process, so I’m relieved to have this live.
ChatGPT: docs.basicmemory.com/integra…
Claude: docs.basicmemory.com/integra…
We hate marketing for Basic Memory, so I made it into a game. This is the first post, so I am winning. the-trail-to-market.fly.dev/
21
Same here, and I’d add that a big upgrade is coming so if you find some features are missing hang tight for the upgrade to v4.0!
You’re unable to view this Post because this account owner limits who can view their Posts. Learn more
Goodbye Tweeps, Hello Twits.
Twitter is now a fundamentally different social network than it was this time last week.
Change the social contract 👉🏻
change participant behavior. 🤡
Fire half the staff 👉🏻
fire on the servers. 🔥
The blue bird is now a black bird.
The conversation for democracy is happening right now. Thanks @madrid_mike for what you do, great stuff!!!
Wow! Biden on 🔥🔥 Join me now on Mike Drop. Ep. 14 BIG movement in the polls Today at 9:30PM ET / 6:30 PM PT. Join us! callin.com/link/hlfCngGPpU
Drew Cain retweeted
As a developer, it’s easy to get fixated on the latest tech.
But, I’ve found quality software can be built in a vast array of languages, libraries, and frameworks.
A strong team with old tools can still deliver amazing results.
Fundamentally it was a communication problem where one team told another team to delete stuff and they deleted the stuff.
atlassian.com/engineering/po…
Drew Cain retweeted
Two decades ago, Silicon Valley startups popularized open floor plans and on-site perks.
Today’s startups have embraced flexibility and remote work. I think this will become the predominant way companies work 10 years from now
Drew Cain retweeted
Software engineering is like 80% writing.
And I'm not talking about code:
- Jira Tickets
- Designs docs
- Presentations
- Emails & slack
- Documentation
- Internal and external blog posts, etc.
You may not realize it, but you try to program humans more than machines.
Drew Cain retweeted
✨ Woo-hoo! You can now export, convert and replay user flows with Cypress! 📹
Try out the new @Cypress_io Chrome Recorder library 👉🏼 goo.gle/cypress-chrome-recor…
Drew Cain retweeted
In which I break down just what @Ubiquiti is alleging against @briankrebs:
lastweekinaws.com/blog/ubiqu…
Drew Cain retweeted
"The resultant javascript one-liner is a single horrendously inscrutable mega-expression." Another remarkable piece of madness from @LingdongH 😍
media.mit.edu/projects/2d-an…