Concept status
The development concept dated 8 May 2026 explores new contextual workflows, chat over events, and agents that generate or test scenarios. It is a direction for development, not a release commitment.
Context workflow
The proposed workflow would organise reference documents and operating context for use by analytical functions. Corpus ownership, versioning and access boundaries remain part of the design.
Chat over events
The concept considers natural-language access to event information and analytical summaries. Any implementation must respect role visibility, source provenance and the difference between stored facts and generated interpretation.
Scenario-generation agent
The concept includes assistance with building scenario graphs. Generated graphs would require structural validation, compatible block/model references, representative testing and controlled publication.
Testing agent
The proposed testing agent would support scenario evaluation. A generated evaluation must remain traceable to a reference dataset, the tested versions and explicit quality criteria; it cannot replace acceptance evidence.
Use in project planning
Confirm availability, supported scope and acceptance conditions for the intended delivery. Do not design a required operational process around an unconfirmed roadmap item.
Proposed per-event Context output
The concept proposes Context as a process alongside detection and verification. Its inputs are the event details, media, event type, site, zone and permitted reference corpus. Its purpose is to enrich the result, not silently decide on behalf of the operator. The intended record retains the model, prompt version, sources and generation time.
| Proposed output | Purpose |
|---|---|
| Tags | Find events concerning work at height or moving equipment |
| Context | Describe visible evidence and risk factors |
| References to rules | Show the applicable document passage |
| Interpretation confidence | Identify insufficient evidence |
| Operator recommendations | Suggest what to check in the original video |
Proposed agent testing and authorized publication
The proposed scenario agent clarifies zone type, available models, schedule and output event, then drafts a graph from library blocks and explains its logic. The draft is tested on assets. A responsible administrator reviews quality, resource use and affected cameras before deciding whether to publish; independent agent publication outside this process is not part of the concept.
The testing concept records event counts, confirmation/rejection rates, FP/FN when ground truth exists, processing time, FPS and load. Ambiguous frames remain available for manual review. Baseline and candidate need the same assets and run conditions.
A chat result should state its period and filters and link to original event records. It inherits the user’s permissions. Changing production models, scenarios or settings requires separate confirmation and a permission check.
The proposed progression is context enrichment → conversational analysis → scenario draft → testing agent → controlled approval. Each stage needs data and acceptance criteria; no release dates are assigned here.