Integral vision. Kernel at 16. RenderMan-alike renderer at 20. $1M on Upwork. Now: own language + database + protocol. 20 AI agents do the coding on my arch.elastcode.com/investors CyprusJoined July 2017
@andy_pavlo I built a database engine (Orion DB) and solved indexing differently — the OID is the index. No B-tree, no side structure. Hash a word to a number, insert at that OID, lookup is a point read through a 16-way trie. More on my X if curious.
@Evalyze_ai
Elastcode · Own compiler, database, protocol, HTTP server & GUI framework — 9 products ship on it, 20 AI agents build it daily, zero dependencies · Cyprus EU (open to relocate) · Developer tools + infrastructure · Pre-seed · Looking for a technical angel who reads code, not just decks
elastcode.com/investors
@splinedrive Yeah, I want to add the support of RISC-V in my platform. The spec is ready. As soon as we finish the self-compilation system, we will do such support
@lightsilver323 Yes, it's exactly why I'm developing my platform. It's not a usual startup but things for the future. If you are interested in such things check my page:
elastcode.com/investors
We didn't want to build a git host. We were forced to.
Every AI coding agent we run needs to read, branch, commit and be audited — and every existing host treats an agent as a human with a stolen password. Rate limits, no per-agent identity, no way to ask "what did this agent actually change, and why did it think that." GitLab and Gitea are built for people. We needed something built for the thing that now writes most of our code.
So we built OrnGit. The decision that made it interesting:
Every repository is its own Orion database.
Not rows in a shared table. One repo = one .orn file. Repo 37 in org 2 lives in git_2_37.orn — a complete, self-contained database.
Orion isn't only a database. It behaves like a filesystem you can query: hierarchical objects, collections and subcollections, binary blobs, transactions — reachable both as a tree and through SQL-like statements. A git repository is already a content-addressed object store with a tree on top. That's not a bad fit for a relational schema. It's the same shape.
What that buys: we can experiment with storage per repository instead of per platform. Different packing, different indexing, different retention — try it on one repo, measure, keep it or throw it away. Nobody running a shared table can do that.
It also produced my favourite bug of the year.
Commits started hanging forever. Not slow — forever. A verb thread parked in CeWaitForEvent and never came back. gdb showed 59 waiters queued on one condition variable.
The cause: transaction completion matched waiters by transaction number ALONE. With one database that's fine. We now had the main DSM database and hundreds of per-repo .orn files open at once — and transaction numbers collide across databases. A commit finishing in one repo was waking a waiter in a completely different one. The thread that should have been woken waited forever.
The architecture we chose caused it. The architecture we chose is also why we could fix it in one place.
274 repositories running on it now.
But storage isn't the point. This is:
AIRecords.
An AI reads the repository and attaches what it thinks — to the repo, to a file, to a function, to a single commit. Each repo carries its own analysis prompt. New commits get scanned against it. The records are browsable and searchable, and they live in the repository, next to the code they describe.
Not chat history. Not a wiki nobody updates. Memory attached to code, at the granularity of code.
Because the real problem with agents isn't that they can't write code. It's that everything they learned about your codebase evaporates when the session ends, and the next one rediscovers it from scratch.
We've added a lot to OrnGit. AIRecords is the reason it exists.
@bakbergenov_1 Agents don't differ from humans a lot. Standard engineering practice. A separate branch then PR. Every agent has own name and knows its duties. AIRecords tell exactly such things. We always use "a coordinator" agent or two. It's a role and every AI worker can take it.
539 Followers 607 FollowingAgencies → Startups → Consultancy
4+ Years | 3 Niches
Building Things That Scale.
Building: https://t.co/q1RGOO9yXr
https://t.co/VTSX7uZ0ob
230 Followers 884 FollowingFounder of https://t.co/3x0yRtLy4f. Career in Data Science.
Four-Leaf is the job search assistant with every tool you’d need to land your next role
6K Followers 7K FollowingFounder & CEO Inference Analytics, building AI for sensitive data industries, AI entrepreneur. Alum: IBM (Big Data), Informatica, MicroStrategy, McKinsey.
352 Followers 164 FollowingSwitch your phone to airplane mode, leaving only space for the scenery. Freedom isn't about going wherever you want, but about being able to stop whenever you w
54 Followers 109 FollowingI am excited about artificial intelligence, technological advancements, education, and customer experience. Ex-OpenAI, currently at AI21 Labs.
79 Followers 359 FollowingSoftware engineer building production systems and AI agents. Interested in databases, distributed systems, developer tools, research, and product engineering.
75 Followers 321 FollowingHead of Product at https://t.co/FHuT0x1hSp, co-founder at https://t.co/Q5On0O0OqG. I like to make things interesting. Dad, designer, code enthusiast.
49K Followers 1K FollowingI do computers. Married to @radkat. 3 kids. Xoogler. @golang team '10~'20. Made LiveJournal, memcached. Now @Tailscale. Mostly in bluer skies these days.
27K Followers 728 FollowingSW Engineer: Kubernetes, GKE, Google Cloud. Dad. Foodie. Star Wars nerd. Expanse fan.
@thock.in on bluesky.
No longer on xitter.
5K Followers 1K FollowingI make open source dev tools at https://t.co/lrm2L4TahM like mise, aube, fnox, hk, pitchfork. Previously I wrote the heroku cli and oclif. ex-amazon/meta.
37K Followers 20K Following#AR #VR developer, consultant, blogger.
I love #gamedev, #China, and the Zergs' #QueenOfBlades.
Most of the time I've no idea what I'm doing.
7K Followers 168 FollowingGraphics researcher. Known for moment shadow maps, blue noise, spectra, light sampling. Opinions are my own.
https://t.co/Q8asoCmY0s
2K Followers 159 FollowingBacking technical founders building the future of AI software, infrastructure & intelligence. Email is best: [email protected]