Skip to content

Vector AI — Flight Diagnostic Assistant

Vector is LogHat's built-in AI diagnostic assistant. It is available on every analyzed flight and is pre-loaded with the full context of your specific log: health score, telemetry summary, battery data, vibration measurements, GPS behavior, EKF status, and the complete forensic report data.

Unlike general-purpose AI assistants, Vector has direct knowledge of the flight you are reviewing. You can ask it specific questions about your log without having to describe the data yourself.

Accessing Vector

Vector is available in the flight analysis view for any completed flight. Open a flight from the Logs page and locate the chat panel. It is loaded alongside the 3D replay and telemetry graphs.

Vector also works without a log. You can ask any general drone or ArduPilot question from the chat panel — no flight required. See Freeform Mode below.

How Vector AI Works

Vector uses a retrieval-augmented generation (RAG) architecture. When you ask a question, Vector does not guess or rely on general knowledge — it retrieves specific measurements, timestamps, and findings from your log's telemetry summary, the AI health score output, and the structured data in the forensic report.

This means Vector's answers are grounded in your actual flight numbers. If you ask about battery sag, the response will reference the specific voltage curves recorded in your log, not a general description of battery behavior. If you ask about EKF issues, the answer will cite the specific timestamps at which lane switches or variance spikes occurred.

Vector is not a general-purpose chatbot. It is scoped to one thing: helping you understand the flight you have open in front of you. There is no hallucination risk on log-specific facts because responses are derived from verified, structured log data — not generated from a language model's parametric memory.

What Vector Can Answer

Vector is designed to respond to diagnostic questions about your flight. Examples of effective questions:

  • "Was the battery healthy during this flight?"
  • "Were there any GPS issues?"
  • "What caused the warning in this flight's health score?"
  • "Did the EKF have any problems?"
  • "What was the maximum vibration on the Z axis?"
  • "When did the flight mode change?"
  • "Was there any risk of a crash based on this log?"
  • "Summarize the key findings from this flight."
  • "How does this flight compare to previous flights by this drone?"

Intent-Based Responses

Vector is designed to recognize the intent behind your question and tailor its response accordingly:

| Question Type | Vector Focus | |---|---| | Battery / voltage questions | Voltage curves, sag, mAh consumed, endurance | | Date / time questions | Precise timestamps from the log | | EKF / error / alert questions | Critical alerts with timestamps | | Vibration questions | Axis-specific data, motor balance, propeller advice | | GPS / satellite questions | Satellite count trends, HDOP, environment factors | | Altitude questions | Climb/descent analysis | | Summary / overview | Comprehensive flight review |

Freeform Mode

You do not need an open flight to use Vector. In freeform mode, Vector draws on its ArduPilot and drone knowledge base to answer general questions without any log context.

Examples:

  • "What does EKF variance mean in ArduPilot?"
  • "What causes toilet bowl effect on a GPS-assisted copter?"
  • "How do I calibrate a compass on a Pixhawk?"
  • "What is the difference between EKF2 and EKF3?"
  • "What pre-arm checks are most commonly failed?"

To use freeform mode, open the Vector chat panel without selecting a specific log, or simply ask a general question when no log is loaded. Vector will automatically respond from its knowledge base rather than from flight telemetry.

Thinking Steps

When Vector processes your question, it shows a brief thinking panel before the response appears. This shows you what Vector is doing in the background:

  1. Knowledge base searched — relevant ArduPilot documentation and telemetry patterns retrieved
  2. Analyzing flight telemetry — log-specific data cross-referenced with your question
  3. Generating expert response — answer being composed

These steps appear progressively before the text response starts streaming. They are informational only and do not require any action from you.

Conversation Memory

Vector retains context across your conversation. You do not need to repeat earlier information in follow-up questions — Vector will refer to the conversation history when formulating responses.

Examples of follow-up questions that work because of conversation memory:

  • "You mentioned the EKF variance spike — what exactly caused it?"
  • "How do I fix that?"
  • "Did this happen on previous flights?"

The conversation context is maintained for the duration of your session. It resets when you navigate away from the flight analysis view.

Example Conversations

EKF Health

You: "Were there EKF issues during this flight?"

Vector: "Yes — the log recorded one EKF lane switch at approximately 4 minutes 22 seconds into the flight, coinciding with a spike in velocity variance on the Y axis. The primary EKF lane failed a consistency check and fell back to lane 2 for approximately 18 seconds before recovering. This is consistent with a brief GPS multipath event or compass heading disturbance. The health score applied a 20-point deduction for this. Recommendation: check the GPS antenna for clear sky view and verify compass calibration before the next flight."

Battery Health

You: "How was the battery health on this flight?"

Vector: "Battery health was acceptable but worth monitoring. The pack started at 24.8V (6S, fully charged) and finished at 21.9V. Voltage sag under peak throttle measured 0.8V total (0.13V per cell), which is within normal range for a healthy 6S pack at moderate loads. Total capacity consumed: 3,240 mAh. Estimated remaining capacity at landing: approximately 18%. One minor low-voltage event was logged at 8 minutes 41 seconds — voltage briefly touched 21.4V during an aggressive climb maneuver. No failsafe was triggered. If sag increases beyond 0.2V per cell under normal load in future flights, consider a load test on the pack."

Vibration Analysis

You: "Was vibration normal during this flight?"

Vector: "Vibration was within normal range across all three axes. Peak readings: VibeX 8.2 m/s², VibeY 9.1 m/s², VibeZ 11.4 m/s². All values are below the 15 m/s² warning threshold. IMU clip count: 0. The Z-axis reading is slightly higher than X and Y, which is common for multirotors with soft-mounted flight controllers — the vertical compliance absorbs horizontal vibration better than vertical. No action is required, but if Z-axis vibration climbs above 15 m/s² in future flights, inspect propeller balance and motor mounting hardware."

What Vector Cannot Do

  • Vector cannot modify your flight settings or parameters remotely.
  • Vector cannot access data from other users' flights.
  • Vector cannot make regulatory determinations or certify airworthiness.
  • Vector's responses are based on the recorded telemetry and are advisory in nature. All operational decisions remain with the pilot-in-command.

Fleet History in Context

If you have multiple flights from the same drone in your LogHat account, Vector can access the health history of that drone across recent flights. This allows it to identify trends — for example: "Vibration on this drone has been increasing across the last four flights, which may indicate a propeller or motor bearing issue developing."

Fleet history context is available only for registered users with multiple completed flights from the same drone.

For Developers — API Access

Vector AI is available via a REST API for external integrations. The API supports:

  • Log-grounded chat — upload a .bin log and ask questions about it
  • Freeform queries — ask general drone/ArduPilot questions without a log
  • OpenAI-compatible endpoint — drop-in compatible with the OpenAI Python SDK, LangChain, and LlamaIndex

Full documentation is available in the Vector AI API Specification. Contact support@loghat.app to obtain an API key.

Data Privacy

Vector AI queries are not stored beyond your active session. Your questions and the responses Vector generates are not used to train or fine-tune any model. All retrieval context is scoped to your specific log — Vector cannot access another user's data, and no cross-user data access is possible at any layer of the system.

Return to Documentation Home

Ready to analyze your first log? See pricing →

Previous

Forensic PDF Report

Next

Supported Platforms