Published: 2026-10-29T23:37:20.792Z

Compatibility: Platform >= 0.2.84; MES >= 0.1.54; Edge Manager >= 0.2.78

Predictive Maintenance priority board, retained signal comparisons and visual review, embedded investigations, unified follow-up and quiet monitoring. Install with MES 0.1.54 on Platform 0.2.84 / Edge 0.2.78. Saved-detail latency and full-shift soak remain qualification limits.

Published: 2026-10-29T23:27:42.826Z

Compatibility: Platform >= 0.2.84; MES >= 0.1.54; Edge Manager >= 0.2.78

Predictive Maintenance priority board, retained signal comparisons and visual review, embedded investigations, unified follow-up and quiet monitoring. Install with MES 0.1.54 on Platform 0.2.84 / Edge 0.2.78. Saved-detail latency and full-shift soak remain qualification limits.

Published: 2026-10-27T17:00:42.187Z

Compatibility: 0.2.83 to -

Persistent reports, automatic context compaction, reviewed PDF and Excel exports, recoverable finalization, machine workbenches, fleet triage and accountable maintenance follow-up. Wide PDF tables retain row identifiers across column groups. Current source authorization, spending allowances and reviewed maintenance actions remain enforced. Requires MES 0.1.52 and Edge Manager 0.2.77. Both native architectures passed qualification.

Published: 2026-10-24T19:33:11.536Z

Compatibility: 0.2.78 to -

Predictive Maintenance reports focus on findings and next actions, with specific source context in place of generic AI disclaimers. Older production exports present recorded totals while preserving the original saved report. Evidence validation, scoped tools and reviewed maintenance actions remain enforced. Requires MES 0.1.45 and Edge Manager 0.2.69. Both native architectures passed qualification.

Published: 2026-10-24T11:37:10.127Z

Compatibility: 0.2.78 to -

Integrated SCADA page, module and project design with screenshot review, connected navigation and retained multi-page context. Explicit continuation recovers accepted drafts after context-limit failures. Predictive Maintenance, scoped tools and evidence validation remain available. Requires DevicePlatform 0.2.78 with Edge Manager 0.2.75 for designer image review. Native ARM64 and AMD64 qualification passed.

Published: 2026-10-05T13:51:48.126Z

Compatibility: 0.2.71 to -

Predictive Maintenance reports focus on findings and next actions, with specific source context in place of generic AI disclaimers. Older production exports present recorded totals while preserving the original saved report. Evidence validation, scoped tools and reviewed maintenance actions remain enforced. Requires MES 0.1.45 and Edge Manager 0.2.69. Both native architectures passed qualification.

Published: 2026-10-04T22:27:05.892Z

Compatibility: 0.2.71 to -

Selected-document questions retain their objective and use validated source passages. Typed production claims preserve owner totals, units and uncertainty. Investigations retain evidence through bounded synthesis and syntax repair. History, case titles, local readiness and reviewed usefulness/cost metrics improve operator follow-up. Includes offline quality evaluation; native demo qualification does not establish real-machine predictive accuracy. Requires MES 0.1.44 and Edge Manager 0.2.69. Offline extraction, scoped SDK 2 tools and administrator-approved grants remain required. Both native architectures passed qualification.

Published: 2026-10-04T17:16:36.819Z

Compatibility: 0.2.71 to -

Saved conversational chats and maintenance cases with follow-up questions, attachments, findings, verification and handovers. Larger bounded investigations reserve capacity for synthesis and preserve partial evidence. Lifetime work compaction, ERP maintenance and part-change evidence, replacement-aware baselines and on-demand historical PromQL improve coverage. Offline extraction, scoped SDK 2 tools and administrator-approved grants remain required. Both native architectures passed qualification.

Published: 2026-10-04T16:57:43.877Z

Compatibility: 0.2.71 to -

Saved conversational chats and maintenance cases with follow-up questions, attachments, findings, verification and handovers. Larger bounded investigations reserve capacity for synthesis and preserve partial evidence. Lifetime work compaction, ERP maintenance and part-change evidence, replacement-aware baselines and on-demand historical PromQL improve coverage. Offline extraction, scoped SDK 2 tools and administrator-approved grants remain required. Both native architectures passed qualification.

Published: 2026-10-04T15:06:18.036Z

Compatibility: 0.2.71 to -

Saved conversational chats and maintenance cases with follow-up questions, attachments, findings, verification and handovers. Larger bounded investigations reserve capacity for synthesis and preserve partial evidence. Lifetime work compaction, ERP maintenance and part-change evidence, replacement-aware baselines and on-demand historical PromQL improve coverage. Offline extraction, scoped SDK 2 tools and administrator-approved grants remain required. Both native architectures passed qualification.

Published: 2026-10-03T22:01:20.879Z

Compatibility: 0.2.70 to -

Saved conversational chats and maintenance cases with follow-up questions, attachments, findings, verification and handovers. Larger bounded investigations reserve capacity for synthesis and preserve partial evidence. Lifetime work compaction, ERP maintenance and part-change evidence, replacement-aware baselines and on-demand historical PromQL improve coverage. Offline extraction, scoped SDK 2 tools and administrator-approved grants remain required. Both native architectures passed qualification.

Published: 2026-10-01T05:28:28.409Z

Compatibility: 0.2.68 to -

Assistant chat now returns one bounded plain reply without source browsing or report generation. Machine investigations remain explicitly selected. Production tools explain and locally validate exact raw owner aggregate fields; exact evidence copies preserve source provenance while catalogs remain definitions, not observations. Authorized saved-job details include the original question and plain answer. Current scope authorization metadata avoids unnecessary owner data extraction during report reads. Strict grounding, paid-request reconciliation, offline isolation, machine/source grants and investigation budgets remain unchanged. Native ARM64 and AMD64 offline-engine qualification passed.

Published: 2026-10-01T03:47:46.858Z

Compatibility: 0.2.68 to -

Longer grounded investigations use a bounded 30-minute execution window and 128,000 cumulative input tokens, retaining conservative UTF-8 admission and the existing 12-turn, 40-tool and 24,000-output limits. Authorized saved-job retry renews configuration-sized access without replaying settled calls or resetting recorded usage. Exact current document revisions are rechecked in settled batches of two before inference; revoked, missing or changed sources stop further reads. Typed elapsed-time recovery is distinct from local request timeout and runtime outage. Full local evidence, raw paid trace, source permissions, offline isolation and consent remain unchanged. Complete native ARM64 and AMD64 offline-engine qualification passed; no calibrated failure probability or remaining-life model is claimed.

Published: 2026-10-01T03:04:43.259Z

Compatibility: 0.2.68 to -

Grounded investigations advance from host-supplied catalog and scoped context inventories into useful manual, operational and analysis sources. Trusted phase guidance and retained assistant progress prevent repeated cold discovery; every refresh tool remains available through bounded typed next-turn activation. Readiness uses fresh app-approved and actor-scoped machine intersections with temporary grants, including authenticated Edge proxy access. Numeric grounding recognizes only complete document/evidence identifiers present in cited sources, preserving measurement validation. Full evidence, raw paid traces, offline isolation, consent and cumulative budgets remain unchanged. Complete native ARM64 and AMD64 offline-engine qualification passed; no calibrated failure probability or remaining-life model is claimed.

Published: 2026-10-01T02:27:42.760Z

Compatibility: 0.2.68 to -

Bounded investigative working memory keeps full local evidence and paid-call audit while exposing exact source observations, manual procedures and citations in scoped source journals. Search snippets retire only after the same immutable extraction was read; source versions, hashes, anchors, quality, review flags and omission counts remain explicit. Maintenance uses a bounded working toolbox with typed local discovery and a final scoped source check. Rejected final interpretations retire only from explicit repair dispatch; permission, export, tool and cumulative inference budgets remain unchanged. HTML/PDF exports show the examined UTC period separately from generation. Complete native ARM64 and AMD64 offline-engine qualification passed; screening priorities are not calibrated failure probabilities or remaining useful life.

Published: 2026-10-01T01:08:23.022Z

Compatibility: 0.2.68 to -

Evidence-grounded investigations retain complete assistant/tool protocol and an independent resumable tool queue. Repeated authorized inventories and condition metadata use bounded model views with useful machine details. A reversible dispatch-only JSON encoding shares identical values within their export class and tabulates repetitive structures without removing numeric observations, manual anchors, scope or caveats; full evidence and messages remain local. Conservative inference budgets, offline isolation and consent remain unchanged. Includes measured maintenance screening, scoped historian/Mongo correlations, approved manuals, local document indexing and durable MCP investigations. Complete native ARM64 and AMD64 offline-engine qualification passed; no calibrated failure probability or remaining-life model is claimed.

Published: 2026-09-30T23:55:59.893Z

Compatibility: 0.2.68 to -

Responsive investigation history and selected artifacts use bounded current owner checks, exact approved-manual revisions and operation/state permissions without cross-request authorization caching. Reconciled paid failures account authoritative usage once and release grants without replay. Direct scoped MCP tools share explicit bounded local investigation contexts across harnesses for collection, deterministic analysis and evidence reads, with actor/source validation and cancellation/restart fencing; they never invoke hosted AI. Preserves local-only document indexing, measured predictive-maintenance screening, PromQL/Mongo correlations and grounded technician checks. Full native ARM64 and AMD64 offline engine qualification passed; no calibrated failure probability or remaining-life model is claimed.

Published: 2026-09-30T23:08:37.198Z

Compatibility: 0.2.68 to -

Document indexing completes locally without hosted inference, returning anchored section counts and quality/review status; retries safely reuse currently authorized exact-version/checksum manual extraction. Paginated compact investigation/report history performs request-local current machine and source checks while opening an artifact retains full checks. Structured bounded tool previews retain complete local evidence and unchanged inference budgets. Accurate temporary-capacity and reconciled-charge recovery actions preserve paid replay protection. Includes measured predictive-maintenance screening, scoped PromQL/Mongo correlations and technician checks. Full native ARM64 and AMD64 offline engine qualification passed; no calibrated failure probability or remaining-life model is claimed.

Published: 2026-09-30T22:42:24.691Z

Compatibility: 0.2.68 to -

Reliable scoped job approval and document processing on ARM Boxes. Investigations wait for durable Edge access approval and preserve cancellation while approval is pending. Mandatory malware scanning has a bounded 180-second cold-start budget inside a 300-second document profile with the same offline, CPU and memory limits. Unexpired actor-owned manual extraction is reused only after current source authorization, exact version/checksum and attachment-byte validation. Includes the evidence-grounded predictive-maintenance condition screen, PromQL/Mongo correlations and source-cited technician checks from 1.0.3. Complete native ARM64 and AMD64 offline document qualification passed.

Published: 2026-09-30T22:24:14.585Z

Compatibility: 0.2.68 to -

Evidence-driven predictive-maintenance investigations: comparable product/run baselines, bounded scoped PromQL, captured stop variables, scrap associations, ranked technician inspections and charted source evidence. Accurate CP3B offline-worker readiness, hosted allowance diagnostics, grounded action instructions, and condition-screen HTML/PDF exports. All bundled document engines passed native ARM64 and AMD64 offline qualification. Screening priorities are descriptive; no calibrated failure probability or remaining-life model is claimed.

Published: 2026-09-29T08:54:40.230Z

Compatibility: 0.2.67 to -

Initial qualified SDK 2 release: durable local investigations, scoped MES and PromQL tools, evidence-based reports, offline document extraction and OCR, administrator-owned schedules, host artifact quotas and fixed-model hosted or self-hosted inference. Ambiguous historical continuous units are excluded from quantity correlations. Both native architectures passed the offline engine probes.