When we shipped Maestro Flight Review as a single-file tool, the next thing we wanted operators to be able to do was drop their whole shift in — every flight from a multi-drone wildfire patrol, all at once. That works now. Drop one log or fifty: you get a per-flight forensics report for every one, plus a shift-level pattern report at the top that surfaces issues no single flight log would show.

The thing about a real shift

If you patrol high-risk zones with a fleet, the most important questions usually aren't "what happened on this flight?" They're "is this airframe underperforming consistently?" and "why does this drone always have GPS trouble over the south ridge?" Single-file forensics can't answer those. The pattern lives in the relationships between flights — the same airframe across many sorties, the same drone over the same patrol area, the same crew working the same weather window. You have to look at the whole shift together.

The open-source PX4 Flight Review web tool is single-file by design, and the usual command-line parsers work on one log at a time. The cloud logbook services will upload everything for you, but they charge a monthly fee and your logs leave your laptop. Most operators end up writing their own scripts to roll a multi-flight comparison — and that's a recipe for "I'll get to it next week" forever.

How it works

Open the Maestro Mac app, hit Cmd+Shift+F to open Flight Review, and drop your logs in (or pick them in the file dialog — multi-select is on). Up to 100 in one pass.

Per-flight analysis. Each log parses locally and is sent — as a sanitised JSON summary, never the raw binary log — to whichever LLM you've configured with your own Anthropic, OpenAI, Mistral, or Google key. The result cards stream in as each flight finishes, so you're not waiting on the whole batch.

If you drop fifty logs, the analyses run one after another rather than all at once. That's deliberate: most operators have rate limits on their API key, and a parallel burst would trip them. One-at-a-time with a visible progress trail is steadier and easier to trust.

Shift-level pattern report. Once two or more flights have finished, an orange-accented "Fleet Analysis" card appears at the top of the panel. The model receives compressed versions of every flight summary — position traces dropped, battery and GPS series thinned, logged messages kept — so an entire shift fits inside the context window, along with a dedicated forensics prompt that asks one question explicitly: "What patterns emerge across these flights that wouldn't be visible in any single one?"

The kinds of findings this is designed to surface:

  • "This drone had GPS dropouts on two flights — both over the south ridge of the patrol area. Treat that sector as RF-degraded and route this airframe's coverage around it."
  • "This airframe's battery margin tightened progressively across three flights — a cell-drift signature. Pull the pack for inspection before the next shift."
  • "All three drones showed elevated wind compensation through the mid-afternoon window, consistent with the gust forecast. Environmental, not an airframe fault."

That third kind of finding is often the most useful. It's the sort of thing an experienced fleet lead would catch only after staring at a dozen plots — separating an environmental window from a real airframe problem — and the model gets to it in seconds.

Chat across the whole shift

The chat panel at the bottom is no longer scoped to a single log. It sees every loaded flight. Useful prompts:

  • "Which airframe had recurring battery issues?"
  • "Show me every flight that triggered an automatic return-to-launch, and why."
  • "What's the wind-exposure pattern across the afternoon flights?"
  • "Compare these two flights — did they cover the same area?"

And if mid-conversation you realise you also want the morning flights in the picture, hit + Files next to the chat input. Drop them in, they parse, and the chat sees them on the next turn.

What this looks like for a wildfire patrol operation

For fire-and-rescue and public-safety crews running multi-drone shifts over defined high-risk zones — early detection of ignition under the operator's own authorizations, not flying into an active fire:

End-of-shift debrief. Pull every log off the cards, drop the folder in, and get a shift report plus per-flight reports in the time it takes to make tea. Each per-flight card has an audit-signed printable report for any flight that needs filing.

Incident review. Something didn't go to plan on one flight. Drop the whole shift in and ask the chat "what was different about that flight versus the others?" The model can compare wind, GPS quality, battery profile, mode-change patterns, and log-message density across all of them, and tell you what's genuinely anomalous instead of what merely looked alarming in the moment.

Predictive maintenance. Every analysed flight is saved into your local fleet history (we go deeper on this in a follow-up post). After enough flights on a given airframe, Maestro starts surfacing degrading-metric trends and statistical outliers measured against that airframe's own historical baseline. No subscription, no cloud, free.

Scale and where your data goes

You can load up to 100 logs per session. The compressed summaries are sized to keep an entire shift well within the model's context window, and the per-file log-size limit is unchanged.

The LLM tokens are billed by whichever provider you've chosen — paid directly to them, never to us. The cheapest configured model handles routine forensics comfortably; switch to a more capable model in Settings when you want deeper reasoning on a tricky flight. Either way, the raw log never leaves your machine — only the sanitised summary is sent, and only to the endpoint you've pointed Maestro at.

Model choice, configurable per provider

Settings has dedicated fields for the model name on every cloud provider — Anthropic, Mistral, OpenAI, Google. Leave a field blank to use the default, or override it to whatever you prefer. Operators on enterprise plans with specific model availability can point Maestro straight at their entitled models, no rebuild required.

Get the build

Free download for Mac: Download Maestro Flight Review (Apple-signed app for Apple Silicon — no email gate, no account).

If you've been waiting to drop your whole shift in, this is the post that says you can. New to Maestro Flight Review? The product page has the full feature list — interactive plots, a 2D map, a time-synced scrubber, rules-based health checks, fleet history with trend detection, A/B compare, audit-signed printable reports, and a vision-based sanity check. All free, all local-first, all under your LLM provider's control.

Related reading: Introducing Maestro Flight Review (the launch post); Predictive Maintenance for PX4 Fleets (how the local flight-history archive surfaces airframe-level trends and outliers); Getting an LLM API key (the practical guide if you've never set one up).