01
Efficient inference
Sparse, local computation designed to reduce unnecessary memory and compute overhead.
AI Research & Engineering · MalaysiaBuilding intelligence
Building intelligence
beyond brute-force scaling.
JustMyth develops efficient, auditable reasoning systems designed from first principles, with a focus on sparse computation, structured memory, local-first inference and rigorous evaluation.
Products
Agents that run on your machine, under your rules.
Our research ships as software. Two desktop products, both pre-alpha, both built around the same idea: capable AI you can inspect, interrupt and run locally.
Myth
aarxiv.org · sparse…
Find the three newest papers on sparse attention and compare them
ShieldGood evening
Open the top results in the background and summarise what each one claims.
On it. Working in 3 background tabs. You can keep browsing.
Ask, or type a URL…Ollama · local
MythBrowse
Pre-alphaAI-first desktop browser
A browser built on the Servo engine (no Chromium, all Rust) where AI chat is the home screen and an agent can work through background tabs while you keep browsing.
- Address bar doubles as an AI prompt
- Any model API or local Ollama
- Built-in ad & tracker shield
Mythworking in backgroundStop
Task
Reconcile October invoices
- Open invoices foldercheckpoint 01
- Read 7 PDF invoicescheckpoint 02
- Update expense sheet in Excelcheckpoint 03
- Email summary to accountantwaiting for you
Guardian · approval needed
Send an email with 1 attachment to your accountant?
Allow onceDenyalso sent to Telegram
brain · APIeyes · localguardian · on
MythDesk
Pre-alphaLocal-first personal AI agent
Myth controls your real Windows PC (apps, files, logged-in sessions) in the background, asks before anything risky, and checkpoints every step so it never loses progress.
- Pull the plug, it still finishes
- Approval rules for risky actions
- Talk to it from your phone
PROJECT Z/01Advanced AI Research
Intelligence should not end at a prompt.
Beyond the prompt.
Project Z/01 is an experimental AI research initiative exploring systems that can reason beyond isolated prompts, adapt through experience, and operate with greater continuity and accountability. Our ambition is to investigate new foundations for intelligence that is efficient, reliable, and guided by verifiable evidence.
These are research objectives, not proven capabilities.
Explore Project Z/01 →In active research · Not publicly available
What we are building
Less brute force.
More structure.
We are exploring systems that use computation selectively, preserve inspectable internal structure and can be evaluated against explicit evidence instead of polished outputs alone.
02
Structured reasoning
Representations and reasoning mechanisms built for inspectability, constraints and verification.
03
Local-first intelligence
Research toward capable systems that can operate on ordinary hardware with lower infrastructure dependence.
04
Evaluation & safety
Clear boundaries between generation, evaluation, evidence and promotion to reduce false confidence.
Current stage
Active R&D: turning architectural hypotheses into implementations, adversarial tests and reproducible evidence.
- Focus
- Efficient reasoning systems
- Approach
- First-principles engineering
- Base
- Malaysia
Frontier-assisted development
Use frontier models as instruments, not authority.
Frontier AI systems including Claude support architecture review, implementation, adversarial testing, scientific auditing and structured experimentation. Claims still have to survive evidence, reproducibility and independent evaluation.
Early access
Get early builds as they become usable.
Join the waitlist for MythBrowse, MythDesk or both. Or, for research, collaboration and company inquiries, get in touch.
- MythBrowseAI-first desktop browser
- MythDeskLocal-first personal AI agent