Matt (@autoglia) header
Matt (@autoglia) profile picture
Joined February 2026
Pinned Tweet
Autoglia is open source (MIT). The goal: make AI agents that remember like people do, not just store messages, but understand who you are, what you're building, and pick up where they left off. Not another memory feature. The actual infrastructure layer that makes persistent agents possible. Open source because the problem is bigger than one product. The more minds on it, the better. Repo is live. Come build with us. autoglia.com
1
59
AI has a memory problem. Not storage — everyone solves that. But LLMs store everything equally. A fact from yesterday and a fact from last year weigh the same. That's not how human memory works. We remember what mattered.
1
51
If memory knows you, it can watch for you. Not reactive — proactive. When something changes that matters. When patterns break. An extension of your awareness, not a database.
1
36
We're still figuring this out. Testing whether this actually works. Building the technical system to find out. Sharing what we learn as we go.
19
It is taking a surprisingly long time to get openclaw via minimax 2.7 to understand how to control my yamaha stereo receiver. The web interface is really throwing it. Once it gets it at least the procedure can be recorded in the db.
52
Just launched v2.13.0 -- now we have memory that learns & benchmarks to understand progress.
1
37
The test scenarios: Startup continuity — does it remember who you are? Project memory — does it know what you're building? Task continuity — does it remember what you were doing? Preference persistence — does it remember your settings? Staleness resistance — does it prioritize fresh over old? Bundle quality — does it return the right mix of context? Pathway learning — does it get better at retrieval over time?
1
25
Why it matters: We run this benchmark after every change. Every new feature either improves the score or it doesn't. No hand-waving. No "trust us."
1
12
The real goal: Make memory quality measurable. So we can make it better. If you're building AI agents, you deserve to know: does your memory actually work?
10
Went from v1.0 to v2.12 in two weeks. Learning system, retrieval scoring, pathway tracking, relation penalties, benchmark tests. The memory gets smarter every time we use it.
1
98
Built a memory system that learns what matters. Not just storing facts -- it tracks which retrievals helped and which got corrected. Over time, it gets better at remembering what you actually need. SQLite. Local-only. That's what 'memory that runs itself' means.
19
A lot of builders are solving recall. Fewer are solving recovery. Those are different problems. Recall = finding something old. Recovery = helping the agent pick up where it left off.
21
Be aware of what's placed in your memory.
AI SEO is now a thing
20