Getting an LLM API Key for Maestro Flight Review

Maestro Flight Review is free because you pay for the LLM, not us. The product asks you to bring your own API key from whichever provider you trust — Anthropic, Mistral, OpenAI, or Google — and routes the AI analysis through that key. Your logs stay on your machine; we never see them; you control the model, the cost, and the data residency. This post is the practical guide to actually getting one of those keys set up if you've never done it before.

You only need one key. Pick whichever you're most comfortable with, paste it into Settings, restart Maestro, drop a .ulg in. The cost for a typical 8-minute flight is between €0.001 and €0.05 depending on which model you choose. A whole shift of twelve sorties: well under €1.

Which provider should I pick?

Our honest take, in priority order:

Anthropic Claude — best forensics analysis in our testing. Strong reasoning, good at citing log messages and parameter names accurately, low hallucination rate on technical content. Default tier (claude-sonnet-4-6) costs about €0.005 per flight; the bigger claude-opus-4-5 costs ~10× more but writes noticeably deeper reports for incident review.
Mistral AI — France-domiciled, GDPR-aligned, EU-data-residency. Pick this if your organisation has procurement constraints about where data is processed. Quality is good; not quite Claude-tier on technical reasoning but consistently better than mid-tier alternatives.
OpenAI GPT — solid baseline, lots of model options. The default gpt-4o-mini is cheap and capable; gpt-4o or gpt-5 if you want to spend more for a better report.
Google Gemini — generous free tier on gemini-2.0-flash and gemini-2.5-pro. Useful if you want to test the workflow without paying anything; production quality is a bit behind Claude / GPT on technical content but it's free.

Anthropic Claude — step-by-step

Go to console.anthropic.com
Sign up with your email (or sign in if you have an account)
Add billing — credit card, minimum top-up around $5. Anthropic bills per-token, so this is a deposit not a subscription. Once it's loaded, your key spends down against it.
Navigate to API Keys in the left nav
Click Create Key, give it a name like "Maestro Flight Review", copy the key (starts with sk-ant-...) — you only see it once, so paste it somewhere safe immediately
Open Maestro Settings (Cmd+,), paste the key into the Anthropic API Key field, click Save, restart Maestro when prompted

Total time: ~5 minutes. The $5 deposit covers roughly 1,000 flight analyses with claude-sonnet-4-6. Set a low monthly spend limit in Anthropic's billing settings if you want a hard ceiling.

Mistral AI — step-by-step (the EU-data-residency choice)

Go to console.mistral.ai
Sign up, verify your email
Add billing — Mistral also has a free tier with rate limits, but production use needs billing enabled
Navigate to API Keys, create one named for Maestro, copy it
Paste into Maestro Settings → Mistral AI API Key, save, restart

Mistral is the right pick if your organisation has a policy that all data processing happens within the EU. Mistral is France-domiciled, contracted under French law, and their infrastructure runs in EU data centres. Combined with Maestro's local-first parsing (the raw .ulg never leaves your machine) you have a fully EU-data-residency-compliant analysis stack.

OpenAI — step-by-step

Go to platform.openai.com
Sign up, add billing (credit card, minimum top-up around $5)
Settings → API keys → Create new secret key
Copy the key (starts with sk-...), paste into Maestro Settings → OpenAI API Key, save, restart

OpenAI's default model in Maestro is gpt-4o-mini — fast and cheap. Override the model in Settings if you want gpt-4o or gpt-5 for deeper analysis. Set a monthly usage limit in OpenAI's billing settings if you want a spend ceiling.

Google Gemini — step-by-step (the free option)

Go to aistudio.google.com and sign in with your Google account
Click Get API key in the left nav, then Create API key
Copy the key (starts with AI...), paste into Maestro Settings → Google Gemini API Key, save, restart

Google's free tier is generous: gemini-2.0-flash gives you 15 requests per minute and 1500 per day at no charge. More than enough for casual / low-volume operators. If you upgrade to a paid Google Cloud account, you get higher rate limits and access to gemini-2.5-pro for deeper analysis.

What does it cost per flight?

Rough numbers for an 8-minute flight summary (~10 KB structured JSON to the model, ~2,000-token markdown report back):

Provider / Model	Per flight (€, ~)	Per 50-sortie shift (€, ~)
Anthropic claude-haiku-4-5	€0.001	€0.05
Anthropic claude-sonnet-4-6 (default)	€0.005	€0.25
Anthropic claude-opus-4-5	€0.05	€2.50
OpenAI gpt-4o-mini (default)	€0.001	€0.05
OpenAI gpt-4o	€0.02	€1.00
Mistral mistral-large-latest (default)	€0.005	€0.25
Mistral mistral-small-latest	€0.0008	€0.04
Google gemini-2.0-flash (default; free tier)	€0.00 *	€0.00 *
Google gemini-2.5-pro	€0.01	€0.50

* Free up to Gemini's daily rate limits (currently 1500 requests/day on gemini-2.0-flash). Above that, paid tier kicks in. Numbers above are rough — provider pricing changes; check the provider's pricing page for current figures.

The fleet-level analysis runs one extra call per session at roughly 2-3× the per-flight cost, so a 50-sortie shift at claude-sonnet-4-6 totals ~€0.30 in tokens. Compare to AirData UAV at $7-99/month flat or DroneLogbook at $5-50/month — BYO LLM is dramatically cheaper for the same workload.

Configure it in Maestro

Open the Maestro Mac app
Settings: Cmd+, or Menu → Maestro → Settings…
Paste your API key into the matching field (Anthropic / Mistral / OpenAI / Google)
Optionally override the model name in the model-name field below the key — leave blank to use the default
Click Save Settings
Choose Restart Now on the prompt so the new key takes effect
After restart, hit Cmd+Shift+F to open Flight Review and drop a .ulg in. The provider badge in each file card should now show the provider's label (CLAUDE / GPT / MISTRAL / GEMINI).

You can configure multiple keys — Maestro will use them in cascade order (Anthropic → Mistral → OpenAI → Gemini, first available wins). This is useful if one provider hits a rate limit or returns an error; Maestro automatically falls through to the next.

Privacy: what actually leaves your machine

To be specific about what gets sent where:

The raw .ulg file: never leaves your machine. Parsing is local.
The structured JSON summary (the output of the parser — mode timeline, battery curve, log messages, etc.): sent over HTTPS to your chosen LLM provider as part of the analysis request.
The model's response: comes back to Maestro, gets stored in your local fleet history (if archiving is on).
Nothing else. No telemetry, no analytics, no "we improve our models with your data" by default. Read your chosen provider's data-use policy — most have opt-out for training data; some (Anthropic, Mistral) don't train on API data at all.

Get the build

Free download for Mac: Download Maestro Flight Review (Apple-signed App for Apple Silicon).