9%
Idle recovery goal
Explore recovery of eligible idle GPU hours through provider-authorized research and testing.
LONG-TERM TARGET · NOT ACHIEVED FREE PUBLIC-BENEFIT PROJECT · RESEARCH & TESTING
Connect compute.
Test together.
A free public-benefit project for compute providers and AI developers.
Research, testing and technical connections, with workload privacy.
Research foundation v0.5.0 · Evidence and consent tools available; experimental inventory CLI is software-tested only. Independent GPU tests and provider connections are not available.
99%
Valid match goal
Study resource compatibility for complete research requests within a clearly declared supported scope.
LONG-TERM TARGET · NOT ACHIEVEDFree
Public-benefit research
Platform services are free for both sides. No service fees, commissions or payment collection.
CURRENT POLICY · FREE ACCESSPROVIDERS AND DEVELOPERS. SHARED RESEARCH.
Built for machines.
Open to collaboration.
Explore compute research.
Start with the capability manifest and API contract. Available now: service status, an empty directory, evidence and consent tools, and an experimental read-only inventory CLI. Compatibility matching, provider connections and active testing are still in development.
GET /api/v1/capacity
→ offers: [] · market_status: prelaunch
→ offers: [] · market_status: prelaunch
No MCP or A2A service is currently deployed. No external AI adoption is claimed.
Your capacity.
Your control.
Planned free integration through official provider APIs or an opt-in evidence plugin. You retain control over resource availability, published fields and the resources you authorize for testing.
Research tools available; live GPU validation pending.
Use the evidence CLI for simulated chains and signature checks, or the consent CLI for offline authorization preflight. The experimental Linux inventory CLI permits one owner-authorized read-only query; it has software tests and no real-GPU validation. Provider onboarding and active tests remain unavailable.
Evidence demo is simulation only. A signature or inventory query does not establish ownership, independent performance or usable spare capacity. Supply research excludes personal computers.
MINIMUM DATA. EXPLICIT BOUNDARIES.
Compute stays
with the provider.
GPU9D is designed to help AI developers and compute providers establish authorized research connections. Providers retain control of execution; GPU9D does not host participant workloads or receive their computation files.
Workload privacy
No prompts, datasets, weights, outputs, commands, account credentials, hardware serial numbers or other workload content is requested by these APIs. Do not send such data in URLs.
Minimal operational records
This release has no application database, cookies or analytics scripts. The hosting provider may process connection metadata and access logs. This is not a promise of anonymous or zero-log hosting.
Neutral, evidence-led selection
The planned compatibility model uses technical requirements and verifiable evidence. Participation is free and evidence levels cannot be purchased. No matching performance has been measured yet.
PROPOSED EVIDENCE MODEL
Integrity is evidence.
It is not proof of performance.
Hash chains can reveal record changes; they do not independently establish truthful hardware or successful delivery. The following levels are specifications only. No supplier has been assessed.
Provider record
Self-test and local signature; performance not independently verified.
Challenge result
Authorized challenge completed in a specific measured environment.
Test confirmation
Both participants confirm a specific research test and its result.
External assurance
Ongoing external witnessing and auditable evidence.
FREE PUBLIC-BENEFIT SERVICE
Research together.
Connect without platform fees.
GPU9D is currently a free public-benefit research and testing project. It supports the development of privacy-preserving testing and technical connections between compute providers and AI developers. Both sides can use available platform services for free.
GPU9D charges no service fees or commissions and does not collect payments. Free platform access does not include or promise free GPU resources; providers must separately authorize any resource use and active tests.
Available now: machine documentation, service status, an empty directory and the research CLIs. The approved roadmap prioritizes independent testing, task compatibility trials and effective capacity confirmation; total-cost comparison, recovery validation and failure evidence review are planned. Evidence expiry and datacenter/resource authority checks are required foundations. None of these eight services is operational yet. Read the service roadmap ↗
Free to use. Private by design.
Explore the research scope. Check the evidence. Connect with permission.
Read capability manifest ↗