--- title: SpinMem: LLM Inference Data Placement (HBM, CXL, NVMe) | SpinForge url: https://spin-forge.com/spinmem description: KV-cache and memory-tier placement software, supplied with a service interface and a vLLM integration hook. --- [← The collection](https://spin-forge.com/products) 03 / KV-cache & memory placement # SpinMem ## Everything in its right place. KV-cache and memory-tier placement software, supplied with a service interface and a vLLM integration hook. See what it solves ↔ Drag to change perspective Customer outcome [Run AI more efficiently](https://spin-forge.com/platform/ai-serving) Available now CLI MCP Headless Choose your access mode THE PRODUCT, IN PRACTICE ## Place across the hierarchy. Keep inference moving. SpinMem treats data placement across L2, HBM, CXL and NVMe as one supported optimization problem. It helps a team evaluate where inference data belongs rather than treating each tier’s settings independently. The supplied service interface and vLLM hook connect placement decisions to the serving workflow. BUILT FOR LLM-serving teams, vLLM operators and inference-infrastructure engineers managing cache across memory tiers. CHALLENGES PRODUCT RESPONSE ## The problem you face. What the product changes. YOUR CHALLENGE WITH SPINMEM 01 ### Fast memory is limited The cache competing for the fastest tier can exceed the capacity available to a serving workload. SpinMem evaluates placement across the tiers represented in the deployment model, giving your system a deliberate allocation rather than an isolated infrastructure setting. 02 ### Placement choices are disconnected Rules chosen separately for different tiers can make the overall placement difficult to assess. The application considers the supported placement problem together, so the team can review one placement decision against the modeled constraints and objective. 03 ### A placement result never reaches serving A useful plan is not enough when an engineer must manually translate it into the runtime. The service interface and vLLM hook connect the placement decision to a supported serving workflow. Integration is focused on where the cache belongs, not on replacing the entire inference stack. WHAT YOU GET ## A defined capability. A useful result. 01 ### Memory-hierarchy placement Placement decisions across the supported L2, HBM, CXL and NVMe model. 02 ### A service interface The supplied FastAPI service for accessing the placement capability. 03 ### A serving connection The supplied vLLM hook for the qualified serving environment. 04 ### A defined deployment scope The modeled tiers, constraints and placement objective against which the application is evaluated. AVAILABLE NOW ## One product. Three ways to work. SpinMem is fully available with CLI, MCP and headless access. Use the interface that fits your terminal, your agents or your automated workflow. 01 / CLI ### For your terminal. Run the product from a terminal or script, alongside the tools your team already uses. [Get CLI access](https://spin-forge.com/contact?product=spinmem&interface=cli&brief=I+would+like+SpinMem+access+through+CLI.+Please+share+the+setup+documentation+and+configuration+for+the+selected+product.+We+will+confirm+the+environment%2C+license+scope+and+workflow+during+onboarding.) 02 / MCP ### For your agents. Connect MCP-compatible clients and agents to the product’s exposed capabilities. [Get MCP access](https://spin-forge.com/contact?product=spinmem&interface=mcp&brief=I+would+like+SpinMem+access+through+MCP.+Please+share+the+setup+documentation+and+configuration+for+the+selected+product.+We+will+confirm+the+environment%2C+license+scope+and+workflow+during+onboarding.) 03 / HEADLESS ### For your workflows. Operate without a graphical interface in unattended, scheduled or pipeline-driven work. [Set up headless access](https://spin-forge.com/contact?product=spinmem&interface=headless&brief=I+would+like+SpinMem+access+through+headless.+Please+share+the+setup+documentation+and+configuration+for+the+selected+product.+We+will+confirm+the+environment%2C+license+scope+and+workflow+during+onboarding.) Product-specific inputs, hardware requirements and result guarantees still apply. An access mode does not change your license or deployment scope. [Explore all product interfaces](https://spin-forge.com/developers?section=access-directory) EXAMPLE CUSTOMER WORKFLOW ## Place cache across an existing memory hierarchy A vLLM team wants to evaluate cache placement before committing to more infrastructure. Illustrative customer use—not a new benchmark or completed customer deployment. 1. 01 Describe the supported tiers, capacities and cache-placement workload. 2. 02 Connect the placement service to the qualified serving path. 3. 03 Compare memory use and serving behavior with the existing placement policy. THE WORK, IN MOTION ## Placement is a decision. Explore the product workflow, from your input to a result you can use. 01 / Inventory KV-cache blocks compete for space across the available memory tiers. Interactive explanation. This illustration does not execute customer workloads or display measured performance. FIT & ACCEPTANCE ## Start with the task. Agree the useful outcome. What you bring The supported inference workload, data and tier model covering L2, HBM, CXL and NVMe, available hardware and acceptance requirements. What a useful evaluation measures - + Tier utilization and capacity pressure - + Serving latency including movement overhead - + Useful concurrency at the accepted quality and service level Delivery License the selected capability, commission an implementation, or agree an operated service. Infrastructure and additional engineering are specified separately. QUESTIONS BEFORE BUYING ## A clear scope. A clear relationship. **What is the important boundary? − **The product formulates supported placement choices as a QUBO. It is distinct from SpinCache representation/direct decode and KX MemoryForge offload/restoration. The best modeled placement is not a universal end-to-end speed or global-optimality guarantee. **How is it priced? + **$90,000/year is the retained standalone price proposal for its named scope. The platform plans buy qualified workflows and capacity, not a second charge every time a compatible method is selected. Existing bundle inclusions do not expand automatically. **How do we get access? + **Fully available with CLI, MCP and headless access. Runtime requirements, supported operations and license scope remain product-specific. ANNUAL APPLICATION LICENSE $90,000 / year One named production deployment. Shared engine rights for this application included. [Scope & packages](https://spin-forge.com/licensing?tab=software) PUT SPINMEM TO WORK ## Tell us what needs to improve. Bring a representative workload and the result your team needs. We’ll define the product, deployment and engineering scope around that task. [Discuss this workflow](https://spin-forge.com/contact?product=spinmem&brief=Product%3A+SpinMem%0AOur+challenge%3A+Fast+memory+is+limited%0AWorkload+%2F+baseline%3A+%0AEnvironment%3A+%0ARequired+output%3A+%0AAcceptance+measures%3A+Tier+utilization+and+capacity+pressure%3B+Serving+latency+including+movement+overhead%3B+Useful+concurrency+at+the+accepted+quality+and+service+level%0ADelivery+preference+%28software+%2F+engineering+%2F+managed%29%3A+) **Related products, workflows and system connections + **RELATED CAPABILITIES ## Connect the right products. Distinct responsibilities. Explore each scope before choosing a combined deployment. [### SpinCache KV-cache compression and direct-decode software for supported long-context LLM serving.](https://spin-forge.com/spincache) [### KX MemoryForge A multi-tier LLM KV-cache runtime for compressing, offloading and restoring cache in a supported memory and storage environment.](https://spin-forge.com/kx-memoryforge) [### SpinProof AI-performance assessment software that profiles a workload, prioritizes optimization changes and checks their effect on model quality.](https://spin-forge.com/spinproof) ONE PLATFORM / OR A DIRECT PURCHASE ## AI Serving with SpinMem. Activate this capability inside a qualified ai serving workflow. Compatible methods and required shared checking are included once within that workflow; the product remains independently purchasable. Capacity, release, environment, result contract and operating responsibilities are specified before activation. Choosing an additional method is not permission to add an unrelated workload or unqualified deployment. [Explore AI Serving](https://spin-forge.com/platform/ai-serving) [Choose workflow capacity](https://spin-forge.com/plans?configuration=ai-serving&product=spinmem) CONNECTED WORKFLOWS ## Use SpinMem in your workflow. The product remains independently available. These customer-specific solution paths use the relevant supported capabilities, with integration and delivery rights agreed separately. [05 ### Performance & release checks Make performance and quality evaluation a repeatable part of your release process. Compare an agreed workload before a change is accepted into production.](https://spin-forge.com/solutions/release-checks) [07 ### Cloud & infrastructure enablement Build a supported optimization, inference-performance or agent-workload offering on your infrastructure, with a clear product boundary and operating model.](https://spin-forge.com/solutions/cloud-provider-services) [Explore all 17 solution workflows](https://spin-forge.com/workflows) FROM PRODUCT TO CUSTOMER WORKFLOW ## A useful outcome. A defined delivery. Commission the model, integration and acceptance work needed for your task. Existing products stay independently available. [03 / Scoped customer program ### Improve a computational workflow Commission a bounded performance improvement with source identity, numerical acceptance and the appropriate checking evidence.](https://spin-forge.com/programs/compute-improvement) [Explore decision programs](https://spin-forge.com/decision-systems) THE WHOLE-SYSTEM ROLE ## SpinMem in a wider workflow. Price retention and placement using future reuse and avoided recomputation, not only bytes occupied. POTENTIAL SYSTEM CONTRIBUTION KV compatibility and placement contracts; savings measured including movement. [S1 · Joint workload & machine optimization](https://spin-forge.com/system-synergies/joint-configuration) [S3 · Checked computational memory](https://spin-forge.com/system-synergies/checked-computational-memory) [See this product in the map](https://spin-forge.com/system-synergies/product-map?q=SpinMem) [Define an integration evaluation](https://spin-forge.com/system-synergies/evaluation) The product is independently available through CLI, MCP and headless access. The joins above are integration candidates; compatibility, authority, data rights and end-to-end benefit must be established for the selected workflow. START WITH THIS PRODUCT Use the software directly or ask us to scope the implementation. [Software access](https://spin-forge.com/plans?product=spinmem) [Have us do the work](https://spin-forge.com/contact?product=spinmem&delivery=engineering)