The first beta of PlumeSentinel AI — version 26.04 — is now available. This release is a minimal evaluation build, intentionally scoped to the smallest end-to-end loop that exercises the core agent workflow.
What’s in 26.04 beta
- Single-agent design. The full project vision is multi-agent; the beta runs a single monitoring agent end-to-end so each piece of the loop — task selection, run orchestration, multimodal reasoning over satellite imagery, structured bulletin generation, telemetry — can be evaluated in isolation before the multi-agent expansion.
- Repeated sweeps over historical wildfire events. The beta works through a manifest of historical wildfire smoke events (the 2020 West Coast outbreak: September 9–11). Three event dates × two sensor configurations = six tasks per sweep. Repeated sweeps support iterative evaluation against a known reference.
- Lightweight Grafana + InfluxDB deployment. Telemetry — deterministic metrics and rendered bulletins — flows from the agent through MQTT (Message Queuing Telemetry Transport) to InfluxDB and renders in a Grafana dashboard.
- Multimodal reasoning over satellite imagery. Each task produces satellite imagery plots from a forecast-evaluation pipeline; the agent reasons over those images and emits a structured situation bulletin for the event date.
Intended use
26.04 beta is for internal evaluation — shaking down the agent loop, the telemetry path, and the bulletin format against a known event before expanding to live data feeds and the full multi-agent design.
What’s next
Subsequent releases will expand to multi-agent orchestration, live data ingestion, and broader event coverage.