Skip to content

Developer guide

Understand how UCM connects the engine's KV Cache lifecycle to external storage, then locate the interface and implementation needed for your change. Read the architecture and request flow before choosing a development task.

Understand the system

  1. Architecture: engine, Connector and Store responsibilities and the storage layers traversed by data.
  2. How caching works: block matching, reuse conditions and the cost of layered storage.
  3. Request lifecycle: Scheduler/Worker metadata handoff and load/save completion.
  4. PD integration: how cross-request reuse combines with the current request's P/D handoff.

Features and design documents

  • Features: usage-oriented guides for supported capabilities — what the feature does, how to enable it, how to verify the effect.
  • Design documents: in-depth design records behind those features — data structures, algorithms and failure semantics.

Continue by task

Development task Guide
Prepare an editable runtime Build from source
Choose and configure a backend Cache Configuration
Add storage or a Pipeline stage Extend Store
Add runtime measurements Metrics development
Integrate a diagnostic or test tool Toolkit architecture and extension
Change native bandwidth tests Dev sandbox
Submit code or documentation Contribute

For deployment configuration, use the user guide. These pages describe current source behavior; runtime verification requires compatible engines and device environments.