5 commits

Author	SHA1	Message	Date
rcourtman	82c615b3b9	Filter virtual disks from SMART checks to prevent false positives (#1329) ... ZFS zvols (zd*), device-mapper, virtio disks, and other virtual block devices don't support SMART and were being reported as FAILED. Use lsblk JSON metadata to filter by device prefix, transport, subsystem, and vendor/model. Also treat missing smart_status as unknown rather than failed, and ignore UNKNOWN health in Patrol/AI signals.	2026-03-08 22:16:24 +00:00
rcourtman	81ec5c525a	feat(ai): parallelize tool execution and refine knowledge extraction ... - Implement parallel execution for read-only tools in agentic loop - Optimize negative marker summaries to be more informative - Fix memory percentage scaling in query tools - Add derived memory stats (avg/max) to extraction logic - Add explicit fresh data intent detection to bypass knowledge gate - Update associated tests	2026-02-01 00:12:36 +00:00
rcourtman	50ecd62ae6	feat(ai): add extractors for all tool actions and fix predict/extract key mismatches ... Major expansion of the knowledge extraction system with extractors for every tool action, plus critical bug fixes discovered during live testing: Extractors added: - query: topology, health, search, list, config (with secondary cached key) - storage: pools, disk_health, raid, backups, backup_tasks, ceph, ceph_details, snapshots, replication, pbs_jobs, resource_disks - metrics: performance, baselines, physical disks, temperatures - alerts: list, findings (with resource_id in key to prevent collisions) - docker: services, updates, swarm, tasks - kubernetes: clusters, nodes, pods, deployments - pmg: status, mail_stats, queues, spam - patrol: get_findings - exec/read: file reads and command execution Bug fixes from live testing: - Config predict key mismatch: model omits node from input, added secondary config:{resource_id}:cached key for reliable gate matching - Baselines predict key mismatch: predict returned baseline:{id} but extractor stores baselines:queried; simplified predict to always return marker key - Query:get composite ID: resource.ID contained node prefix (delly:minipc:100), now prefers VMID (int) over composite string - Alerts findings wrong JSON field: struct expected "findings" but API returns "active" with separate "counts" object - Finding key collision: two findings with same key but different resource_id overwrote each other; added resource_id to fact key - Backup tasks vmid type: API returns int but struct had string, causing json.Unmarshal failure; also fixed status case-sensitivity (OK vs ok) Tests: 91+ tests including 38 roundtrip cases, 5 gate flow tests, negative marker test. All pass.	2026-01-31 22:24:54 +00:00
rcourtman	82ddeac454	refactor(ai): refine agentic loop compaction and knowledge accumulation ... - Inject wrap-up nudges/escalations after token/turn thresholds are met - Update compaction logic to include key accumulated facts in summaries - Refine knowledge extraction and accumulation tests - Update main entry point for revised AI configuration	2026-01-31 19:33:43 +00:00
rcourtman	c5717d1a10	feat(ai): add knowledge accumulation and enhance agentic loop ... - Introduce KnowledgeAccumulator to persist facts across turns - Enhance AgenticLoop to support knowledge injection and final text summaries - Update chat service to wire up knowledge components - Frontend updates to support enhanced chat capabilities	2026-01-31 16:22:50 +00:00