Recent changes (Latest docs)

This page tracks Latest documentation against the current UMKIM v0.2 line. The 0.2.0 doc snapshot is refreshed from Latest via scripts/sync-latest-to-020.py (sidebar in apps/docs/src/content/versions/0.2.0.json). 0.1.0 remains the legacy DCM-era freeze.

Fleet LLM and GPU inference

• LiteLLM hub image pin v1.102.2 (fleet_llm_image_tag in roles/fleet_llm/defaults/main.yml).
• Load balancing across spokes with the same model alias; deployment_affinity disabled (API-key pinning); session_affinity via x-litellm-session-id / metadata.session_id (Redis).
• Auto backends from Ready llama-cpp-helm FleetApps; umkim llm backends sync-from-apps.
• FleetApp --llm-backend / annotation fleet.umkim.com/llm-backend.
• Fleet Llama profiles CRD, console Apps → Llama profiles, umkim llm profile *.
• Verification: verify-litellm-routing.sh, verify-fleet-llm.sh, verify-fleet-llama-profile.sh, verify-llm-inference-e2e.sh.

FleetApp and console

• Install Helm chart UI improvements; LiteLLM registration checkbox; llama profile help.
• fleet-apps docs: app update, rocky-10-vm, profile install paths.

YAIDAP and site provisioning

• Hub inventory store, sync/import, bare-metal vs VM kinds, YAIDAP_RESULT contract.
• umkim site provision yaidap-sync / POST …/provision/yaidap/sync.
• Operator guide — binding, inventory schema, setup checklist.

Install, teardown, verification

• Install the hub and hub lifecycle: default teardown deletes fleet clusters; --skip-cluster-delete for hub-only.
• Verification: LLM and llama profile scripts documented.
• Versions and pins: LiteLLM image pin called out separately from harbor_fleet_image_tag.

Where to read more

Topic	Page
YAIDAP setup	Operator guide
LLM proxy	fleet-llm
Llama profiles	Fleet Llama profiles
CLI	umkim CLI

Canonical component sources: docs/components/ in the UMKIM repository.