What Essence does, in plain terms
Seven short sections, about five minutes. Each ends with one sentence to remember, and terms are explained where they first appear. The longer, visual version is the Visual Primer.
Software fixes the steps in advance
A program is a list of exact steps, written by people for a particular kind of machine. That works until something changes: the machine, the data or the goal. Then people rewrite the steps, which takes time and money. Steps written in advance also cannot take advantage of the machine that is actually in front of them.
Software fixes the steps in advance.
You say what you want. Essence works out how.
Essence reverses the order. A person, or an AI system, states the outcome wanted and the limits that apply: how fast, how cheap, how secure. Essence then works out the steps at the moment the job runs, using the machine that is actually there.
An analogy: a navigation app. You enter the destination, not every turn, and if a road closes the app picks another route. The destination is fixed and the route is not.
This approach is called wantware, because it starts from what is wanted. It is not about writing or generating code as applications. To keep requests precise, Essence describes them with Meaning Coordinates: a fixed, public set of 256 building blocks, so the same request always means the same thing.
You say what you want. Essence works out how.
Nothing runs until it says why
By design, every request has to state who is asking, what it will do and why before it can start. A built-in check confirms that statement against who is allowed to do what. Each run is designed to leave a record, so an auditor can check later what happened and why.
This describes how the platform is designed. Section 7 says what runs today.
Before and while work runs, the system checks why it is running, and keeps a record.
Not AI, and built to work with it
Essence does not use AI to decide or to execute what runs. The same request always resolves to the same precise description, so results can be explained and repeated.
An AI system can connect to Essence and suggest what someone wants. It can only propose. Before anything runs, a built-in check confirms who is asking and whether they are allowed, and each run is designed to leave a record.
That is the answer to a common worry about AI: that it acts on its own, reaches outcomes nobody intended, and leaves no clear account of why. AI may suggest. Rules that people set decide, and the record shows who asked for what and why.
AI can propose. It cannot decide what runs. People and policy do.
Wherever the machine is not known in advance
The measured results so far come from cloud GPUs and one integrated GPU in a university lab. The same idea applies wherever the machine is not known in advance.
Edge computing and physical AI, such as robots and vehicles, run on many different small machines, often with tight limits on power, space and connection. Steps written in advance for one machine do not carry over well. Essence is designed to work out the steps on the machine that is actually there. The company sees these as massive opportunities. Results on edge hardware have not yet been measured.
Wherever the machine is not known in advance, steps written in advance fall short. That includes the edge and physical AI.
What has been measured
The measured results come from one tool, a GPU optimizer, which generates the instructions a GPU runs while the job is running. In pilot tests on cloud GPUs and on one university lab GPU, each job was compared with the first run of the same job.
Approximate range of speedups across all recorded tests, over the first run of the same job.
Approximate typical (median) speedup on the cloud GPUs.
Approximate range of energy reduction recorded across the pilot log.
On cloud NVIDIA GPUs, speedups ranged from about 4× to 54×. Rowan University’s Digital Engineering Hub ran 18 workloads on one integrated AMD GPU, with speedups from about 14× to 114× and energy reductions of about 98 to 99.7%. Rowan’s report is a draft and has not been signed.
The figures come from the company’s pilot log and the Rowan draft report. The AWS runs were recorded independently by PREDICTif Solutions in a project funded by Amazon. The Google Cloud and Oracle Cloud runs are the company’s own. All were on a single GPU, and results depend on the workload. The two ends of the range are single results, and the typical result is the better guide.
Faster than the first run of the same job, measured on one GPU. It is not a promise for every workload. What we guarantee is real time optimization of data ordering, scheduling, memory, networking and hardware parallelization, which is what highly skilled engineers do by hand and cannot do in real time.
What runs today
One tool runs today. Everything else is either a recording or returns when funded work is complete.
Running today
The GPU optimization build at adaptwithchameleon.com, which generates GPU instructions at run time.
First 30 days of funded work
File-based output, a way to hand results to existing tools, is re-established.
When the milestones are complete
The operating-system-hosted runtime, the signal and video processing layer and the other capabilities return and can be deployed again.
Two funded milestones come next. One lets many GPUs in one machine work together. The other lets people create workloads in plain language. Recorded demonstrations exist, some of an earlier version, and each one is labeled.
One tool runs today. The rest returns with the funded milestones.
Essence replaces steps written in advance with a statement of what is wanted, and works out the steps when the job runs. Today one tool demonstrates this, with measured GPU results. The rest of the platform returns when the funded milestones are complete.
Essence is relevant to two kinds of organization: those that run shared computing infrastructure, such as cloud hosting providers, and those that want to optimize a workflow inside their own application or computing environment.
The Deployment section explains how either would adopt Essence alongside the systems it already runs: in what order, one workload, product or team at a time, and through which integration paths, without replacing its platform or disrupting existing workflows. It also marks which stages depend on the funded milestones.