datahike

BuddhiLW 2026-02-06T00:51:33.066329Z

Datahike is now supported backend in https://github.com/hive-agi/hive-mcp. I'm myself using it. Took 6 minutes to (successfully) migrated from datalevin to datahike (my KG Knowledge Graph). Some features nicely fit in what I'm trying to do with low-level abstraction knowledge, like where does things are in the codebase, what are there in the files etc. E.g., a low-level map of the territory. And having snapshots of confidences, as well as some similar good stuff. Thanks!

🚀 1
whilo 2026-02-06T01:04:57.315419Z

This is super nice! I have also an FRP system for concurrent distributed programming with isolation features for agents that I am working with and hope to release soon. I should take a look at hive-mcp. I saw it is AGPL which makes it difficult to get traction in my experience, but I understand the appeal from a pure open source perspective. Btw. I have built a claude code like TUI (but much simpler) on the weekend for it also backed by Datahike for session and fork mangement. It makes it easier to hook into isolated agentic contexts this way, #simmis is similar but as a web frontend:

1
👀 1
BuddhiLW 2026-02-06T02:06:57.369259Z

Yeah, the greatest appeal is to big tech to just not snatch it and don't give it back. Just reap the rewards, yeah. For the power-user, they have a head-start of about 1-6 months in my calculations. Until big tech catches up. I'm building a proprietary solution that leverages hive-mcp. But, I wouldn't be unreasonable in negotiating with companies that are interesting in acquiring a talent (me 🤠) + want to build on top of existing work. Dual licensing is in the horizon. Maybe next release cycle I will update the README, when I figure out the best terms to write the dual licensing agreement transparently, for whoever may be interested. I don't have a big name, nor brand, and believe hardly anyone cares aside if they possible could gain credits for something they didn't envision or built. So, power to the power user, and let's play the game there is to play, to the rest of the board.

BuddhiLW 2026-02-06T02:08:24.889459Z

Believe or not, I have seen one copy-cat from big tech 2 weeks after I released a first version of hive-mcp 😂. Fair enough, though.

BuddhiLW 2026-02-06T02:11:29.443349Z

That's dope. Currently, I'm thinking on how could I scale up to 100's to 1000's of agents working. Emacs is a ceiling. And I don't think terminals will be the way, but headless claudes or openrouter agents with contextual calls stringed somehow (what I called "drones"). Hivemind (talks and interfaces with you) -> lings (claudes doing work, plan with context a mid-task) -> drones (grunt work, edit files etc) [could call up to 10's easily with smart file-locking and diffing and merging logic involved]

BuddhiLW 2026-02-06T02:12:16.988469Z

So, if you can spawn up to 10's of claudes headless, and them call 10's of drones each, potentially on the verge of 100's of agents doing code in a code-base already

whilo 2026-02-06T01:08:07.775849Z

I also just managed to build a persistent columnar index on dtype-next that allows analytic workflows similar to DuckDB. Hope to release this soon, too. Then we will have a comprehensive set of secondary indices which don't break the persistent memory semantics and git-like functionality.

🔥 1