👨👩👦 Family Man 🦾 @kujolang
Michigan, USA
Joined May 2015
- Tweets36.2K
- Following96
- Followers2.6K
- Likes67.2K
One thing that drives me nuts with coding agents is when they finish and claim to be done, without anything to back that up.
And Kujo Spec was built to cure that frustration.
It lets you give the work clear requirements and acceptance criteria up front, so there's something to check against when your agent finishes
Point your agent at the docs and have them set it up
docs.kujolang.ai/tools/spec/
1
15
Just released: RepoRadar v0.5.0 🚀
- Added agent-friendly CLI reviews, JSON reports, and diagnostics.
- Added persistent hidden repositories with restore controls.
- Fixed overflowing detail buttons and long branch labels; added the Tabler chevron.
- Resolved 15 hardening findings covering scans, backups, imports, API security, metadata, and activity history.
Use RepoRadar to keep every local repo in view, so unfinished work doesn't slip through the cracks.
github.com/robertdevore/repo…
1
28
The Kujo ecosystem provides a lot of accountability for agentic workflows, but the language itself does a lot of the heavy lifting 💪
Native AI record/replay, cassettes, and deterministic request hashing mean you can reproduce model responses, debug failures, and run tests without paying for the same inference over and over.
Built into the language in order to provide your AI-native software development with everything it needs to succeed.
kujolang.ai
1
18
Accountability for agentic workflows looks like this.
If you haven't tried Kujo yet, you're missing one of the greatest opportunities of 2027 ... you gotta be thinking about your future right now 🫡
From start to finish, Kujo has you covered for all of your AI-native software development.
Check it out 👀
2
84
Thank you to those of you who reached out already. Exciting times ahead 🫡
You can't stop this train. Best thing to do now is get in early while theres still time.
me@robertdevore.com
If you want to be taken serious, you have to put in the effort.
If not, what's the point?
Kujo is open to investors, and I used the Kujo Presentations tool to build a proper deck for it.
Kujo lang -> Kujo SSG -> Kujo Presentstions -> You.
invest.kujolang.ai
1
5
91
I tested 100 contributors one at a time and concurrently.
The old way took 11m 36s while Source's concurrent batch took 2m 6s
That's about 5.5× faster, and i haven't even truly started working on performance focused gains yet 💪
With this release, i tested 250 concurrent agent processes across 10 repos.
All edits were retained through submission, combined verification, and promotion 🔥
I also ran 100 agents concurrently on a single repo and it handled that fine as well.
Next tests will be 250 on one repo, then 1,000 💪
3
60
Its good to see decentralized AI publishing gaining traction.
Especially in places with people who seemed to be against it a year or two ago.
Growth is good 🫡
3
70
With this release, i tested 250 concurrent agent processes across 10 repos.
All edits were retained through submission, combined verification, and promotion 🔥
I also ran 100 agents concurrently on a single repo and it handled that fine as well.
Next tests will be 250 on one repo, then 1,000 💪
Source 0.2.0 is out.
This release expands Source's workflow for coordinating contributions from humans and agents: assemble proposed changes, verify the combined result, keep evidence, and promote what was reviewed.
What's new:
- Candidate integration: Combine compatible proposals using real Git merge semantics, with conflict and dependency checks and preserved contributor attribution.
- Evidence-bound promotion: Verification and approval apply to the exact candidate being promoted. Stale or mismatched evidence cannot authorize a different revision.
- Durable verification: PostgreSQL-backed workers support queued checks, deduplication, recovery, and evidence reuse when its validity conditions still hold.
- GitHub coexistence: Keep GitHub as origin and add Source as a secondary remote. source sync copies local branches and tags atomically without force-pushing or deleting remote refs.
- Recoverable submissions: Authenticated, keyed PostgreSQL submissions can recover their original result after crashes or lost responses without creating duplicate contributions.
We also exercised Source's own repository through isolated component tests, browser staging, promotion, worker crash recovery, and coordinated backup and restore.
Source is built in JavaScript on Node.js, with integrations across the Kujo ecosystem
github.com/kujolang/source
Also, you don't have to move everything over just to try Source.
Keep GitHub (or whatever) as origin and add Source as a secondary remote.
Sync your local branches and tags without changing your existing workflow, then start trying Source's review and verification features alongside it.
You can build your confidence with a secondary copy before deciding whether to make Source your primary setup 💪
Im doing the same for my repos now
1
22
Strata includes 3 agent skills (Context, Start, and Memory), with 5 memory modes:
- RECALL retrieves relevant context
- OBSERVE tracks potential memories during work
- CONSOLIDATE saves useful long-term knowledge
- REVIEW audits and cleans up existing memories
- AUTO manages the workflow automatically
Works with any model provider, harness, etc. One place for all of your memories.
Imagine the dataset you'll have in a year 😤
Strata holds memory from all of my usage across all accounts github.com/robertdevore/stra…
3
179
🔥 Awesome AI Programming Languages 🔥
A curated list of programming languages, DSLs, and programming models for AI-native software development.
Got one that's not on the list? Add yours and join the fun 🫡
github.com/robertdevore/awes…
Let's talk about Zero vs Kujo
Graph hashes and patch validation are great if you're trying to maintain the integrity of a program's representation.
Zerolang provides mechanisms for handling those changes. But graph integrity doesn't establish that an app works correctly.
Your agent can make perfectly valid changes that introduce the wrong business logic, break an integration, or fail to satisfy the original request.
Kujo's Spec and Eval let you define what the resulting software is supposed to do and checks the selected outcomes.
We all know that AI agents are good at producing changes that *look* reasonable.
But what we also need to be reminded of from time to time is that clean patches can still implement wrong features.
Personally, I'd rather have an agent whose work can be checked against the actual assignment than one whose changes are carelessly accepted by the editing system.
Kujo is built around making those checks an intregal part of the dev process, as they should be.
kujolang.ai/compare
4
78
I've been loving my Kujo Hardening skill lately.
"Use the kujo hardening skill on this repo"
So simple yet insanely useful. Its constantly tightening up the codebase of each and every repo i point it at.
In one pass in one of my repos today it fixed writer shutdown races, storage pagination/scoping, file whitespace loss, and Worker dependency vulnerabilities.
In another, it completed six fixes covering auth, SDK memory/JSON handling, SMTP fixtures, and webhook lease safety.
Try it out and you'll end up like me with an agent who's sole job is code hardening at the end of another agents session
github.com/kujolang/kujo-ski…
Tired of dealing with an archaic CMS, expensive hosting and an endless stream of vulnerabilities?
Meet Website Cloner 👋
Clone public sites locally - no AI or API key required.
Get an editable static site with Markdown, shared templates and your assets, built with Kujo.
Install it, paste a URL and review your clone.
kujolang.ai/ecosystem/websit…
I've tested this with over a dozen different sites of varying content and styles.
It works great when cloning WordPress sites, Astro sites and bespoke builds.
By now we all agree that most websites don't need a full blown CMS to perform its intended duties.
So why expose yourself to bloatware that leaves you vulnerable to attacks because a plugin author (or CMS maintainer) is slop shipping updates "because AI is so good".
Skip the bullshit and ship something clean that removes the attack vectors you've been told are just part of the gig.
The Kujo SSG provides everything your site needs ... content types and taxonomies, rss feed, sitemap.xml, llms.txt, search and social optimized, automatic webp image conversion and so much more.
The Website Cloner separates templates and content with front matter to handle all of the SEO etc.
All flat file structure. All editable by you, or your agents.
In 2026 there's no reason to use the old CMS you're used to. So dont.
Use Website Cloner to free yourself from the old world and step into the future 🫡
3
94
Give your Hermes workflows evidence you can inspect.
Kujo brings engineering tools, human approvals and execution receipts into Hermes. Watch it catch a checkout bug, fix it and rerun the same tests.
Install: github.com/kujolang/kujo-her…
Shout out to @Teknium for merging it so quick after the submission, and thank you to @NousResearch for putting on the contest 🫡🪽
Here's to hoping I win!
the Kujo MCP server just passed 10K initializations 🔥
Point your agent at it and let it learn about Kujo
mcp.kujolang.ai/mcp
3
78
If you build a lot of repos and need a good way to track them, I've got you covered.
RepoRadar is a local git command center that lets you add your local repos and visualize things like the current branch, if there's upstream/downstream commits, the last commit, dependabot alerts, etc.
You can tag and group your repos to help make it easier to keep track of your various projects.
For me, this is a game changer as I work on 80+ repos across the @kujolang ecosystem as well as personal projects, business ventures, etc.
Get it on GitHub 💪
github.com/robertdevore/repo…
1
6
115