How it fits together

STUDIO CSMALLERFASTERMORE CAPABLECompressionrMLXE-RLMHARNESSUSEFUL PRIVATE AIConsumerBusiness

Six questions investors ask

Why does this need to be one company?

Because the pieces compound. Compression makes a 27B fit, rMLX makes it run, the Harness makes it reliable and useful, and E-RLM is the next step in how it thinks. Each attacks a different bottleneck.

Why wouldn’t the big labs just do this?

We don’t compete on weights. We build the layer that takes a model from any provider — DeepSeek, Kimi, Gemma, Qwen — and compresses, recovers, runs and deploys it on hardware people already own, with the product on top.

Where is the business?

Ready to start a closed pilot of the platform with businesses. LocallyAI AU, our subsidiary, installs and supports it: the machine, the on-site install, configuration and support.

What is the primary business?

Local install and configuration of on-device AI for Australian businesses, through trained contractors. Alongside it, the planned LocallyAI Platform: a private machine in our cluster, rented by the month.

Why so many industries?

Two reasons. We put learning first and apply it across the whole stack, so it stays one cohesive product. And the system is built to be adaptable and reliable, so it is genuinely useful across many industries.

Does the compressed model keep its intelligence?

Divergence 0.3737 inside the 0.40 gate, repetition 0.0394 inside 0.25, over 32,736 scored positions. A task-level benchmark against the full-precision model is the next receipt and the release gate.

The business

Primary business activityLocal install and configuration of on-device AI, through remotely trained contractors across Australia, paid per install (subject to change) and sometimes per hour where the job needs it: staff training on the platform, AI workflow courses and the like. The per-hour model is under consideration, not settled. Secondary: the LocallyAI Platform, plannedBusinesses rent a VM or a physical machine in one of our clusters for a flat monthly or yearly fee, theirs alone: a private compute platform and AI inference provider with private search, to trial a machine, keep one, or use one on demand. At the end of a lease they receive a download of their disk that only they can decrypt.
StudioC.ai

Studio C / pre-seed, September 2026

Frontier-capable AI, on ordinary hardware.

One stack, three bottlenecks, one product: capable models moved onto hardware you own.

Shipping nowStudioC Harness: private AI that finds, understands and gets work done. Ready to start a closed pilot of the platform with businesses. Ships with Ornith-1.5-35B-A3B, Qwen3.8-27B, Ornith-1.5-9B and Qwen3.5-9B. Being productisedProject Anneal and rMLX: a 27B model in 5.77 GiB that runs on a laptop, and the Rust runtime to run it fast. Models, planned1B, 4B, 8B, 14B and other sizes for edge devices. Gemma 4 (26B and 31B), Qwen3.8-27B, Qwen3.8 180B and Qwen3-Next 80B-A3B on everyday consumer hardware and phones. DeepSeek V4 Flash, Kimi K3 and other frontier models for non-commercial use, with runtimes for enthusiast and business hardware. Frontier researchE-RLM: the runtime answers the model’s arithmetic and runs its code while it is still thinking.

Three co-founders: Harry, Matt and Riley. Studio C owns the technology; LocallyAI AU installs and supports it.

1 / 04

StudioC.ai

StudioC Harness / local-first workspace

Harness

Private AI that finds, understands and gets work done — on a machine in your building. Nothing leaves unless you switch it on.

300 / 300

structured tool calls with the Harness’s automatic correction, from 287 / 300 as written, on the stock 4-bit model.

Current stageReady to start a closed pilot of the platform with businesses, while we develop workflows and run internal security audits. Open-code release expected in the coming months, free for businesses and consumers to run. Ships withOrnith-1.5-35B-A3B, Qwen3.8-27B, Ornith-1.5-9B and Qwen3.5-9B. Our own compressed models ship once recovery reaches 90% or more; the next slide has where that stands. Installed for youStudio C has established LocallyAI AU as a subsidiary: the Australian approved installer of the Harness, providing the machine, on-site install, configuration and support. LicenceOpen code, not open source, licence to be decided. Free for personal and business use. Building it into your own software needs approval and may carry a fee, as do components and rMLX. Forge stays internal. Secure Core, nextA planned AI-integrated operating system, built from the ground up to be highly secure and integrated well with the onboard models to get more done.

The product / live footage of the interface, no concept renders

Private connectionsStudioC Search: private web search and fetch through an isolated service, off by default. StudioC Mail: Outlook, Gmail and IMAP — read, draft, send, ask. StudioC Connect: a secure route back to the appliance for staff off site. Each enabled explicitly, each with a permanent disclaimer. Security, enforcedRead-only, write-with-approval or full access per chat. Denials beat every group grant. Admin settings behind a two-step edit lock; secrets masked, write-only, never round-tripped.
Industry knowledgeSix Australian packs — legal, accounting & tax, health & safety, employment, medical, property & planning — searched beside your own files. Configure only the packs you need. Meetings and toolsRecord with consent; transcript, summary and owned actions. Ask in chat for a GST calculator or a chart and it appears working, inline, reusable from the library.
As written287 / 300
Auto-corrected300 / 300

Valid structured tool calls, 300 prompts, the stock 4-bit model. The Harness recognises and corrects malformed calls automatically from a code library of malfunctions captured over thousands of tool calls, part of the Forge compress-and-recover pipeline. No weight changed. Also 12 of 12 on an Exercism-based code set.

Scheduled work · live footage

2 / 04

Project Anneal / complete package size, GiB, same Qwen3.8-27B

Same model.
Smaller by 88.5%.

BF16 original50.10
Stock 4-bit14.95
Ridge 3.7-bit11.73
StudioC runtime7.581
StudioC complete5.769

Measured on the 7.58 GiB runtime

Speed15.45 tok/s on an M1 Max against the stock 4-bit build’s 17.13, same process, n=8. We are 9.8% slower. QualityDivergence 0.3737 against a 0.40 gate, pass. Repetition 0.0394 against 0.25, pass. 32,736 scored positions. Memory2.4211 bits a weight, peak 10.565 GiB at a 4,096-token prompt.

Project rMLX / against MLX 0.32.2

G2 graph build3.07×
G3a dispatch1.90×
G3b fused chain7.95×
G4 auto-fused6.48×
G5 fp16 matmul1.00×
G6a decode0.93×

A full Rust rewrite of Apple’s MLX. Parity 174 / 174 first. Paired medians, n=14, one M1 Max, $0 of cloud. Two gates are open: the fused chain misses our own 0.47 µs bar, and decode is 408.5 tok/s against MLX’s 437. Released as open code under the same terms as the Harness.

The 5.769 GiB build is untrained; the 7.581 GiB runtime carries the quality numbers. Ridge’s figure is third-party, from its model card. Every size is the whole package.

StudioC.ai

Pre-seed / September 2026

27B.
5.77 GiB.

8.68× smaller than the original weights. Complete, 1.5-bit ternary.

The only runtime for this model that stays usable on the machine, with room left for the OS and an agentic context window. Retains an estimated 80% of knowledge; we are working to 90% or more before it ships.

A hybrid rounding-encoding method, different to standard quantisation, with a short recovery run (3–24 hours by model size) in our Forge pipeline. It applies to any model. The Harness recovers what compression costs, correcting known malformations at kernel level: this compressed 27B goes from 0 of 15 structured tool calls as written to 15 of 15 with automatic correction. It is not shipped yet; the Harness runs stock models until then. Forge, the engine behind it, has also found a 0.262-bit compression on the MLP blocks it has reached so far and is iterating on it actively; Forge itself is not planned for release.

Coming next

1B · 2 bits4B · 2 bits8B · 1.5 bits14B · 1.5 bits35B-MoE · 1.5 bits80B-MoE · 1.5 bits180B-MoE · 1.5 bits

Also planned: DeepSeek V4 Flash, Kimi K3 and other frontier models compressed for non-commercial use, with runtimes for enthusiast and business hardware; Gemma 4 (26B and 31B) and Qwen3.8-27B, Qwen3.8 180B and Qwen3-Next 80B-A3B on everyday consumer hardware and phones; 1B to 14B and other sizes on edge devices.

Neither the encoding nor the recovery pass needs the original training data. Runtime today: our own; LM Studio and llama.cpp cannot run it yet.

3 / 04

StudioC.ai

Original architecture

E-RLM

Executable Reasoning Language Model.

The runtime answers the model’s arithmetic and runs its code while it is still thinking. No request, no wait, no reply.

Inside the thoughtThe answer lands within the reasoning block, so every later step reads a value the model never had to compute. It is not a tool call. Token efficientThe model spends its tokens on the reasoning and the new arguments. Everything in orange is the runtime’s, produced once and carried forward, never re-typed. Exact, not rightThe arithmetic is exact. A wrong expression is computed faithfully. Calculation error goes; reasoning error stays. How it’s measuredExact final answers on held-out problems, a binding diagnostic for each computed value, and a wrong-value control.

Early research, not in any product yet. Capability numbers stay internal for now.

Two live examples / a person types at 8 tokens a second, the model answers at 15, the runtime fills in orange

4 / 04