Every message gets read. Mention the shape — you'll hear back from Julian, not a form.
Tribunus is shipping one product in 2026 — Tessera Studio, a desktop app for running large language models locally. If you want to help build it, use it, back it, or bring hardware and real workloads to the research, here's how.
You probably fit one of these. Each one says what we're looking for and how to get in touch.
The fastest way to help is to use the product and report what breaks. We ship to early users first. macOS is the live target; Linux is in active development.
If you are willing to install a developer-preview app, run a real model — Llama, Qwen, Gemma — and file issues on the GitHub repo, you are shape 01.
Small, focused PRs, reviewed personally. We do not have a large team — a small list of people trusted to land a clean diff, and we are always looking to add one or two more.
The areas where help is needed right now: SwiftUI on the macOS app, C++/CMake on the Linux build of the inference side, Core ML / ANE if you have shipped a real model to Apple's Neural Engine, and UX / macOS HIG where the surface needs polish.
Tessera is a fork of llama.cpp with per-tensor quantization calibration: each tensor gets settings tuned by a small search, so models run fast and stay accurate.
If you have a hypothesis about per-tensor quantization, mixed-precision attention, or a new drafter model, the contribution surface is the record. Show a result that reproduces, and we can ship something together.
Specifically people who have done one of three things: (a) shipped a Mac desktop product to a paying audience, (b) maintained a serious fork of llama.cpp, or (c) been a technical co-founder of a small, focused, profitable company.
What we offer: equity, attention, and prompt replies. No board seat required — an hour a week and a straight answer when something's wrong. The bar is real shipping, not credentials.
Tribunus is raising a pre-seed. The company is self-funded and shipping: Tessera Studio is a live macOS developer preview, and the engine behind it — Tessera, our C++ fork of llama.cpp — is open source, with calibrated per-tensor quantization and native speculative decoding.
If you back technical founders with shipped code, let's talk.
Five shapes of partnership, each with a clear deliverable. Hardware, datacenter, engineering, research, edge & robotics. The full framing lives on prism-engine.tribunus.dev.
Make a provider legible. We need hardware access, runtime documentation, kernels, topology, and reproducible workloads.
Observe the deployment. We need representative serving requirements, memory and latency constraints, and operational traces.
Build the missing system. We need a clearly scoped compiler, runtime, provider, or integration problem.
Explore what is not known. We need a falsifiable question, an observation plan, and a clear line between research claims and product claims.
Carry intent to the edge. We need device constraints, sensor or control workloads, and failure-mode evidence.
Shared interfaces stay open. Reproducible fixtures and non-sensitive validation artifacts can become public — the evidence class is recorded either way.
Proprietary stays confidential. Models, telemetry, hardware details, and deployment constraints remain private when the engagement requires it.
Deck-only conversations. The code and GitHub are the source of truth — not slides.
"AI wrapper" pitches. Tessera Studio runs models locally — it's not a chat wrapper over a remote API.
Cold-pitch full-time hires. The first full-time hire comes after a pre-seed close and will be posted publicly.
Every message gets read. Mention the shape — you'll hear back from Julian, not a form.