SuperAPI AI Directory

lazycat-aipod-developer

General & other

View repository

Updated 10-16 02:27

About this skill

AI Pod application: resource and device compatibility

First calibrate the two sources

Before writing/modifying configuration or build/deploy/release, read and execute references/spec-sync.md, and confirm the microservice, CLI, and compute pod/consumer-side versions.You must also read the AI Pod independent official topic:

The general LPK source repository does not cover all AI Pod capabilities. The resource import location/projection path in the current topic differs from the general V2 specification; first follow references/aipod-app-spec.md to identify the boundaries; you cannot guess at a conversion and then claim it is deployable.

Action workflow

  1. Read references/aipod-app-spec.md, select agxorin/thor compose according to the user's device, and do not assume all devices are just one kind of Jetson.
  2. New configurations use config/aipod.yml in resource (without an outer aipod), the model directory, extensions, and the platform startup gate; the old ai-pod-service/browser-extensions serve only as a compatibility fallback.
  3. Verify each device's image architecture/GPU software stack; the AI host must be consistent and end with -ai, and join traefik-shared-network; configure a real readiness probe for long-initialization services.
  4. Use LZC_AGENT_DATA_DIR for data, LZC_AGENT_CACHE_DIR for cache, and LZC_SERVICE_ID for the Traefik identifier; the platform defaults to the NVIDIA runtime, so there is no need to mechanically add gpus/runtime.
  5. If the topic is inconsistent with the general schema, verify the target toolchain and resource consumer before generating the complete package; when only doing build/static validation, do not claim real deployment, model inference, or compatibility with all models.

Installation, release, and device changes require authorization; for microservice names, actively execute lzc-cli box default. Do not copy device serial numbers or microservice names from the documentation into artifacts.