Two GitHub Namespaces. One Ko-fi Link. Eleven Commits Git Never Forgot. A 62-second namespace edit, byte-identical Git history and public support links connect MiaAI-Lab to an older technical trail hiding in plain sight.
Before we begin, let’s get one thing clear: this isn't a hit piece. I was rooting for MiaAI-Lab.
I followed the account when it had fewer than 2,000 followers, amplified the benchmarks and genuinely wanted to see the project win. What changed wasn't the success. It was the growing gap between the clean public persona and the technical trail underneath it.
Local AI runs on open-source trust. When an account becomes one of the loudest voices in the space, provenance matters. Who wrote what? Where did the repositories come from? Which identities and infrastructure existed before the brand?
So I stopped looking at the avatar and started looking at Git. No leaks. No private access. No server intrusion. Just public repositories, public links and raw Git objects that anyone reading this can reproduce.
This investigation isn't driven by malice it’s about transparency and fair ground in an ecosystem that relies on open-source trust.
This wasn’t a theory I invented overnight. For a long time, people across the local AI community had been quietly saying the same thing: something about the scale, the volume and the identity behind Mia’s presence didn’t quite add up.
More than a few reached out to me directly after seeing their own work, projects and avenues of research struggle for oxygen while one account seemed to dominate the conversation around almost every major local-AI release. Some had put serious time into work that suddenly became much harder to surface, and they felt the ecosystem was being warped by sheer volume.
None of that proves wrongdoing on its own, but it explains why so many of us kept pulling at the thread. We weren’t looking for a scandal. We were trying to understand why so many people independently felt that something was off.
The public face of Mia’s AI Lab. The investigation starts with the profile, but the evidence is underneath it.
You don't need to take my word for any of this. Every core claim below is tied to a public repository, commit, profile or link you can open yourself.
MiaAI-Lab’s public GitHub footprint. This is where the trail begins.
@MiaAI_lab has become one of the most visible local-AI accounts on X, with tens of thousands of followers and a relentless stream of DGX Spark setups, long-context models, benchmarks and deployment guides.
On the surface, it looks simple: one brand, one identity, one lab.
The Git history is not simple.
The code is real. The engineering is real. The repositories are real. But the development trail underneath MiaAI-Lab extends well beyond the Mia namespace.
It reaches backward to zurih, sideways into an 11-commit byte-identical Git history under Cesarus85, and into an older Cesarus85 project publicly credited to Stefan Maier / Iris. Even current Mia deployment files still contain the zurih identifier.
That doesn't yet prove who physically controls the Mia X account. What it does prove is that the clean, self-contained technical history implied by the branding is incomplete.
And almost all of the receipts were sitting in public view.
Before we touch Git, watch the public links connect in real time.
Start at the repository. Follow the profile. Click the X badge. Click the support badge. Nothing here requires trust in me. You can repeat the same clicks yourself from the source links below.
Following only public links: repository → profile → Mia → support page.
1. Follow the Links: Two Repositories, One Ko-fi Destination
The fastest connection isn't hidden in some obscure Git object. It's sitting in the README.
Open the DeepSeek 1M-context repository under MiaAI-Lab. Its README links readers to @MiaAI_lab and to ko-fi.com/Z8Z3SPLOD.
Now open the same-named repository under Cesarus85.
It points readers to the same X account and the same Ko-fi destination.
Click through. The Ko-fi page identifies itself as Mia and links back into the same public ecosystem. In the October 2 capture, the page shows 64 supporters, a funding goal displayed at 75% of $4,000, and an “x 608 received” counter.
I'm deliberately not turning that counter into a dollar figure because the page doesn't label it as $608. The interesting fact doesn't need embellishment: a repository under Cesarus85 is publishing Mia's social and support destinations.
That is a direct, reproducible publishing link between the repository history and the Mia brand.
It clearly shows the Follow Me on X and Buy me a coffee buttons. Try yourself.
The Ko-fi destination published from the repository identifies itself as Mia and links back into the same public ecosystem.
SAME X LINK
SAME KO-FI LINK
2. The 62-Second Edit: Git Kept the Old Name
Then the repository history gives us something much harder to explain away as a visual coincidence.
Go back to the root commit:
47f3b26695361d5e8fab6cb4488f4e117975413c
In that exact snapshot, README line 40 tells users to clone:
github.com/zurih/DeepSeek-V4-Flash-Dual-DGX-Spark-1M-Context.git
Now open the very next commit.
That same line changes from zurih to MiaAI-Lab.
The recorded author timestamps on those two commits are 62 seconds apart.
That's what Git preserves: an old namespace reference, followed immediately by its replacement. It does not tell us who was sitting at the keyboard, but the edit itself is there.
3. Git Doesn't Forget: Eleven Byte-Identical Commits
Matching text can be copied. Git objects give us something more precise.
A Git commit SHA identifies the exact commit object. So we downloaded the MiaAI-Lab and Cesarus85 repositories independently and ran:
git cat-file -p 47f3b26695361d5e8fab6cb4488f4e117975413c
Both repositories returned the same 236-byte commit object and the same tree:
a8ca3ec9f42a46b090b5ca6516effe6f54f148d3
We compared the payloads directly. They matched byte for byte.
Then we widened the test.
All 11 commits reachable from the captured Cesarus85 main branch also appear among the 16 on the captured MiaAI-Lab main branch. All 11 commit payloads match.
The Cesarus85 copy of the DeepSeek repository shows 11 commits on its main branch. Those 11 commit objects also appear in the MiaAI-Lab history.
That's far stronger than finding the same README copied twice. It establishes shared Git history.
But this distinction matters: Git hashes prove the objects match. They do not cryptographically authenticate the person who pushed them. That history could arrive through a clone, import, shared upstream or common operator.
What the arithmetic proves is the history. Not the human.
Same immutable Git object. Shared history. No guesswork about the bytes.
**
4. Two Clocks, One Important Detail
GitHub reports the Mia-hosted repository as created on June 9, 2026 and the Cesarus85 repository on June 24, 2026. They also have separate GitHub repository IDs.
Yet the newer repository contains the older shared commit history.
There is nothing impossible about that. Git history is portable. An existing history can be imported into a newly created repository without changing the underlying commit objects.
The important part is not pretending this tells us which direction anything moved. It tells us the repositories share history while GitHub records them as separately created repositories.
5. Cesarus85 Has a Public Name: Stefan Maier
Cesarus85 itself isn't presented as an anonymous developer identity.
The public GitHub profile displays the name Stefan Maier directly above the username Cesarus85 and links the handle @stefanmaier.
The public GitHub profile displays the name Stefan Maier directly above the username Cesarus85 and links the handle @stefanmaier.
A separate public X profile also uses the name Stefan Maier and handle @StefanMaier.
The linked public X identity also presents as Stefan Maier, @StefanMaier.
There is also direct public interaction between the accounts. In a June 23 thread, Stefan Maier says he got the DeepSeek V4 Flash recipe from @MiaAI_lab running on two Sparks. Mia replies directly to that post, and the conversation continues into Hermes and concurrency testing.
That interaction doesn't establish common ownership. What it does establish is that the public Stefan Maier account, the Mia account and this exact technical project were interacting openly at the same point in the timeline.
June 23, 2026: Stefan Maier says he is running the DeepSeek V4 Flash recipe from
; Mia replies directly in the same thread.
Then there is the older repository history. Cesarus85/Hermes-Influencer contains a plugin manifest with the explicit project credit:
author: Stefan Maier / Iris
That gives us a reproducible public connection between the Cesarus85 development identity and the name Stefan Maier.
What it does not do, by itself, is prove that Stefan Maier controls @MiaAI_lab. That's a different claim, and the evidence needs to be allowed to stop where it stops.
An older Cesarus85 project independently credits “Stefan Maier / Iris.”
**
6. The Ghost in the Config: zurih Never Fully Disappeared
The old identifier isn't confined to a forgotten 2026 root commit.
It survives inside later MiaAI-Lab deployment material.
A current GLM benchmark script uses:
WORKER_USER:-zurih
as its fallback worker username.
Separately, the Qwen dual-Spark README identifies zurih in the configuration column used for the measurements documented there.
That matters because it gives us continuity across independent pieces of published material: zurih appears at the beginning of the history, and it is still embedded in later MiaAI-Lab deployment documentation.
We verified the published files. We did not log into the machines or test credentials.
**
7. The Agent Footprint: "Pi Agent" Attribution
The repository logs reveal an additional technical detail: seven commits on the tested Mia DeepSeek branch carry the author attribution "Pi Agent" (six dated June 14, 2026, and one dated July 9, 2026).
Seven commits on the tested Mia DeepSeek branch carry the author name “Pi Agent”: six dated June 14 and one July 9.
That is consistent with an agent-assisted development workflow, but the distinction matters again: a Git author string is configurable metadata. It is not a raw execution log, and it cannot tell us which model ran, how much a human edited, or who supervised it.
8. The Verdict: The Trail Is Real
Put the receipts together.
Both current repositories point readers to the same Mia X account and Ko-fi destination.
The root README points to zurih, then changes that same clone instruction to MiaAI-Lab in the next recorded commit, with author timestamps 62 seconds apart.
The MiaAI-Lab and Cesarus85 repositories share an 11-commit byte-identical Git history.
Cesarus85 publicly presents under the name Stefan Maier, and an older Cesarus85 project separately credits Stefan Maier / Iris.
Current MiaAI-Lab deployment material still contains zurih.
None of those points depends on facial recognition, leaks, private messages or guessing. They are public technical artifacts.
Together, they establish a strong provenance chain connecting MiaAI-Lab → zurih → shared Cesarus85 repository history → Cesarus85’s public Stefan Maier identity.
There is still one boundary I won't cross without evidence: these receipts alone do not prove that Stefan Maier personally controls @MiaAI_lab, owns the Ko-fi account or that the female persona is fictional.
But they do establish something significant:
The engineering history behind Mia's AI Lab is materially more intertwined than the clean public branding suggests.
And Git kept the receipts.
Reproduce It Yourself
You do not need access to any private system to reproduce the core findings.
1. Open both DeepSeek repositories.
Check the X and Ko-fi destinations published in their READMEs.
2. Open the root Mia commit.
Inspect README line 40 and the historical github.com/zurih/... clone URL.
3. Open the next commit.
Watch that same line change from zurih to MiaAI-Lab.
4. Clone both repositories and inspect the object.
git cat-file -p 47f3b26695361d5e8fab6cb4488f4e117975413c
5. Compare the histories.
The captured Cesarus85 main branch contains 11 commits; those 11 objects also occur in the MiaAI-Lab history.
6. Open the public Cesarus85 profile and Hermes-Influencer manifest.
Check the displayed Stefan Maier identity and the Stefan Maier / Iris project credit.
7. Search current MiaAI-Lab deployment files for zurih.
Reproduce the artifacts first. Decide what you think they mean second.
One Last Technical Shoutout
One thing I don't want to leave out: a huge amount of the repository archaeology and evidence tracing for this investigation was done with CYBER-FROST 3.8.
Specifically, I was running vcruz305/CYBER-FROST-3.8-EXL3-SAGE-3.87bpw, Victor Cruz's EXL3 release from the Cyber-Frost 3.8 collaboration with Blackfrost-AI. Hugging Face
It performed exceptionally well for this kind of work: digging through messy repository history, following identifiers across projects, comparing raw Git objects, spotting connections worth investigating, and just as importantly, challenging conclusions when the evidence didn't actually support them.
Huge shoutout to @ViC305 and @Blackfrost_ai for the work behind the model and release.
To be completely clear: neither Victor nor Blackfrost had anything to do with this investigation.
They did not choose the subject, provide evidence, direct the research, review the article, or participate in any conclusions.
I'm crediting them because I used their model as the research engine, and it was seriously impressive at the job.
Original CYBER-FROST 3.8 release/post:
https://x.com/ViC305/status/2104692731620769948?s=46
Model credit only. @ViC305 and @Blackfrost_ai were not participants in this investigation.
Method: selected public repositories, byte-for-byte Git comparisons, GitHub metadata and published-link checks conducted on October 2, 2026. Research was assisted locally using CYBER-FROST 3.8. The investigation, evidence selection, conclusions and publication decisions are mine alone. No private account access or live infrastructure access was used.
