Representation and capacity
SpinCache changes the supported KV representation; SpinMem evaluates placement and KX MemoryForge carries supported offload and restoration into execution.
S1 / SYSTEM CONNECTIONExplore all 64 product roles
Change what fits.
Change what is possible.
A representation change can make a larger batch fit. That batch can make a different kernel useful. Routing, placement and resource limits then change the best operating point again. Evaluate the combination around the useful work your customer needs—not a sum of isolated speedups.
Cross-product integration / customer qualificationExisting products · CLI · MCP · Headless
The products retain their available interfaces. These connections describe integration and evaluation opportunities—not a claim that every combined runtime has been delivered.
Token Factory owns a distinct fleet objective.
The existing Token Factory schedules a GPU fleet as a QUBO to maximize tokens per watt. SpinRoute, SpinMem, SpinKernel, SpinSovereign and GridForge retain their own routing, placement, implementation and policy/resource responsibilities.
Explore the eleven products ↗THE CUSTOMER CHALLENGEWhat separate
What separate
tools can miss.
A locally faster component can leave the real bottleneck untouched. A locally slower change can be essential to a better system configuration. One-at-a-time tuning can therefore reject the combination that actually helps.
THE COMBINED APPROACH TO EVALUATEConnect the
Connect the
right responsibilities.
Commission a joint evaluation of compatible cache representations, memory placement, runtime movement, routing, kernels and resource assignments. Select configurations under the same workload, accepted quality and operating limits.
A new feasible workload or operating configuration—not just a better component timing.
HOW THE PRODUCTS RELATEThe useful
The useful
intersection.
Existing components have distinct jobs. The proposed integration connects their responsibilities without treating access interfaces as proof of a shared runtime.
Compute and routing
SpinKernel evaluates compatible configurations and SpinRoute contributes expert-routing choices. Neither substitutes for the other.
Whole-workload planning
Atlas, GridForge and SpinSovereign contribute their respective execution, resource and policy constraints. Select one owner for cross-product budgets and control timescales.
Measured acceptance
SpinProof evaluates complete performance and quality. Relevant solver checks establish only the mathematical properties they actually support.
INTERACTIVE STUDY / NOT A PRODUCT BENCHMARK
See the connection.
A local browser illustration of the idea. No customer model is uploaded and no SpinForge engine is invoked.
Assumed completed-work units = min(memory-limited capacity, compute capacity) − overhead. Baseline capacities: 40/40. Compression raises the first to 80 and costs 2 units; the kernel change raises the second to 80. Equal accepted output quality is stipulated. The units are abstract, not measured requests or tokens.
ASSUMED MODEL OUTPUT
Baseline configuration
Change one or both controls to compare the combinations.
The combined ratio of 1.95× belongs only to these assumptions. It is not a SpinForge speedup, serving result, GPU measurement or guarantee.
WHAT YOU COMMISSIONAn evaluation.
An evaluation.
A defined delivery.
Not another mandatory product license. Scope the model, integration, acceptance and deployment needed for the actual customer workload.
A joint configuration space
List the compatible choices, coupled limits and settings that may change together.
A matched comparison
Compare the baseline, individual changes and combined configurations at the same accepted quality.
A selected deployment scope
Identify the exact products, environment, operating authority, rollout and rollback.
A complete cost record
Include data movement, setup, checking, orchestration and ongoing resource cost.
ILLUSTRATIVE CUSTOMER WORKFLOW
Long-context inference under a memory ceiling
A serving team has useful compute capacity that cannot be reached with its current memory representation. Evaluate whether a compatible cache change, kernel choice and placement together unlock useful concurrency. A customer result must be measured; the illustration below is not that measurement.
ACCEPTANCE MEASURES
- Accepted useful work at fixed quality
- Tail latency and memory footprint
- Complete execution and coordination cost
- Rollout stability and rollback behavior
SCOPE & EVIDENCEClear at
Clear at
every boundary.
Documented components and proposed cross-product work are not the same evidence.
What the source supports
The component roles are described in the supplied portfolio. The cross-product advantage is an integration hypothesis. The capacity example is a local calculation from stated assumptions, not a SpinForge benchmark.
The integration to establish
Shared configuration and cost contracts, compatible runtime adapters, control ownership, and an end-to-end qualification harness.
What must not be assumed
Component availability does not establish the combined deployment. Separate control timescales are required; do not change kernels or physical infrastructure on every token. GPU capacity, quality and resource commitments remain workload-specific.
Products remain independently usable
Choose only the components that fit. Original and upgrade products are not additive speedups. No generalized benefit, hosted activation, certification or execution authority is implied.
SELECTED CAPABILITIESFind the
Find the
product role.
These links identify candidate contributors. They do not mean every product is required or that every connector already exists.
TAKE THE IDEA TO YOUR WORKLOADDefine what
Define what
must improve.
Start with an original model or workload, an accepted baseline and the protected requirements. Agree the comparison before promising a system-level gain.