
Connect
Access models and agent capabilities through a familiar gateway. Bring your task; choose the tools it needs.
Models, agents and compute, woven into one open network. Terza brings a builder-first vision to Bittensor — useful intelligence, with every contribution traceable.
BITTENSOR SUBNET CONCEPT ↗OPEN MODELS. INDEPENDENT COMPUTE. VERIFIABLE CONTRIBUTIONS. ONE NETWORK, BUILT TOGETHER.
One place to build. An open network underneath.
A proposed network where work can be inspected. Follow a request from execution to evaluation and its usage receipt. Select a stage to explore this illustrative flow.
STAGE08 of 30 · Miner
FROMworker.execution · result = completed
STEPinspect the contribution record
STATUSIllustrative flow · no live network connected
A simple idea: builders bring the task, miners do the work, and validators evaluate the result. Useful contributions should be what moves the network forward.
Explore the flow→
Access models and agent capabilities through a familiar gateway. Bring your task; choose the tools it needs.

Independent miners supply inference and execution. Each result carries a record of the work that produced it.

Validators evaluate outputs against the subnet’s task and scoring rules. Mechanism design is still in development.
Explore the proposed developer experience. These examples illustrate the interface and network flow; they do not execute requests or connect to a live subnet.
const response = await client.chat.completions.create({
model: "terza/open-model",
messages: [{ role: "user", content: "Build something useful." }],
stream: true
});
VISUAL STUDY · NOT NETWORK METRICSA network is only as useful as the work it delivers. Our proposed control plane puts execution, evaluation and usage in the same view.
Follow each request through its model, worker and runtime.
Inspect how a contribution is evaluated, with scoring criteria made explicit.
Connect usage back to the request and the agent that initiated it.
The principles behind a proposed Terza subnet. These are architecture goals, not live network statistics.
a design goal: accessible interfaces for independent builders
proposed roles: builders, miners and validators

evaluate outputs against real application tasks
subnet number and registration are not yet decided
A proposed trace from request to output and evaluation.
a research focus, not a measured network benchmark
an exploration of the Terza subnet direction

Intelligence becomes infrastructure when anyone can contribute, every output can be evaluated, and useful work comes first.
A proposed direction for Terza on Bittensor · concept stage
applications that use the network
inference and agent outputs
quality against shared tasks
A decentralized network needs explicit boundaries and inspectable results. These principles guide the design; security implementation is not yet available.
Design goal: each request defines the permitted task, input and output. Agent capabilities should remain explicit and scoped.
Design goal: publish task definitions and make scoring understandable, reproducible and open to review.
Design goal: make data handling and provider selection explicit. Privacy guarantees will depend on the final protocol and implementation.
An architectural map of the proposed subnet: from application requests to useful outputs and independent evaluation.
A proposed entry point for products that need model inference and agent execution.
PROPOSED INTERFACEIndependent operators serving useful model outputs against a shared task specification.
DESIGN IN PROGRESSA proposed execution layer for bounded tasks with explicit inputs and inspectable results.
CONCEPT STAGEEvaluation designs that compare task quality, latency and reliable execution.
UNDER RESEARCHExplore ways to provide relevant context while keeping provenance attached to each request.
PROPOSED CAPABILITYA common record of request, response and evaluation to make network behavior inspectable.
DESIGN GOALA subnet concept within the Bittensor ecosystem. Registration and subnet identity are not yet set.
SUBNET ID · TBDAn approach built around understandable specifications and reproducible integration examples.
PLANNEDExplore a potential role in the network. Participation requirements, incentives and availability remain to be defined.
Explore applications powered by useful intelligence.
Explore serving model and agent task outputs.
Explore independent evaluation of useful work.
A clear view of what is proposed, what is undecided, and what this concept demonstrates.
Explore TerzaNo live subnet is represented by this website. This is a proposed Terza direction for decentralized model inference and agent execution. The subnet number, final positioning and launch plan are still undecided.
The concept connects applications with independent inference and agent operators, then evaluates the usefulness of their outputs. The task specification, API and evaluation methodology remain to be defined.
No. The interface contains illustrative workflows and design goals, not production performance, customers, revenue or token returns.
The proposed roles are to serve task outputs and evaluate them against a common specification. Hardware requirements, registration, scoring and incentives are not yet published.
The playground on this page is a local interface demonstration. It does not submit requests to a live subnet or expose an available API.
Explore Terza for the existing Terza website and the official Bittensor documentation to understand the ecosystem. This concept does not announce an official partnership or a registered subnet.