Open-weight models changed the cost of intelligence. The model itself can be downloaded for free. What still costs money is the machine willing to keep it loaded, answer requests, and stay online.
That is what Reasoning Capture is building: a network of ordinary computers running open models and getting paid when they do useful work.
You install one line in your terminal, download a reasoning node, choose a model your computer can handle, and leave it running. When a request comes in, your machine answers it. You earn in USDC.
On the other side, a person or agent gets cheap access to open-weight models through an API or chat interface. They subscribe, send prompts, and our assignment agent routes each request to a computer that can handle it.
Not every prompt needs the biggest model on earth. Some jobs can run on smaller local models. Some need stronger machines. The network should know the difference.
Idle computers become reasoning nodes. Open models become usable infrastructure. Intelligence gets cheaper for everyone.
Idle hardware should not stay idle
There are Mac Minis, gaming PCs, office desktops, and old workstations sitting around with unused compute. Most of them will never be part of the AI economy because they are not in a data center. We think that is backwards.
If your computer can run a useful model, it should be able to earn.
Reasoning Capture turns that machine into a host. You are not reselling someone else’s API. You are not pretending a browser tab is doing the work. Your computer loads the model. Your computer receives the job. Your computer writes the answer.
The network coordinates the rest.
The assignment agent
When a prompt enters Reasoning Capture, it does not just go anywhere.
Our assignment agent looks at the request, the available machines, the models they are running, their reliability, and their capacity. Then it routes the job to a host that can serve it.
Some machines will run smaller models. Some will run stronger models. Over time, larger rigs can support larger open-weight systems, while consumer hardware handles the huge number of requests that do not require maximum reasoning power.
That is how we keep the network cheap. We do not need every computer to be a supercomputer. We need the right computer for the right job.
Cheap access for people and agents
Buyers get access through a simple subscription, API, or chat interface. The target price is intentionally low. A $10 monthly subscription should be enough for people to start using the network, testing agents, and sending real prompts without thinking about massive cloud bills.
Agents matter even more. Software can call APIs all day. It
can change a baseURL, send requests, and pay
for inference without a human sitting in the window.
Reasoning Capture is being built for that world: cheap,
available intelligence that agents can spend on directly.
Hosts earn in USDC
If you run a node, you should be paid in dollars. That is why host earnings are designed around USDC. A host can choose to receive USDC and never touch the token. The point is simple: if your machine does real work, you should receive real payment.
A computer that is merely awake earns nothing. A computer with a model loaded, connected to the network, passing checks, and answering jobs can earn.
There are two reward streams.
First, when your machine completes a job, you receive a share of that job’s price.
Second, trading fees from RCAP activity can be converted into USDC and distributed to eligible hosts. The team is also acquiring RCAP so part of that can support the reward pool.
Why RCAP exists
RCAP is not the paycheck. RCAP is the network asset.
Subscription revenue can flow into RCAP, creating a direct link between usage of the network and the token. If someone pays $10 per month to use Reasoning Capture, that subscription can go into the chart or a USDC pool tied to the token economy.
Hosts still earn in USDC. But they can also choose to stake RCAP later if they want deeper participation in the network.
Two paths: run hardware and earn dollars, or hold and stake RCAP because you believe the network will grow. You can do one, both, or neither.
Verification matters
A decentralized AI network only works if the machines are honest. That means we cannot just trust every computer that says it answered a job. The coordinator has to meter requests, check outputs, and remove unreliable hosts.
The plan is to re-run a slice of jobs across other machines and compare results. Hosts that pass checks build trust and receive more opportunities. Hosts that fail lose access to work and rewards.
This is what separates a real compute network from a points farm.
The vision
Reasoning Capture is building the cheapest practical inference layer for open-weight AI. Not by owning every GPU. Not by building another giant cloud lab. By connecting machines people already own and routing work intelligently across them.
Old computers, Mac Minis, gaming PCs, office machines, and larger rigs can all become part of the same reasoning network. Small models handle simple jobs. Stronger machines handle harder jobs.
Buyers get cheap AI. Hosts earn USDC. RCAP captures network usage.
That is the loop.