Concepts & glossary

Short definitions of the terms that show up across these docs and the product.

Investigation

Any AI-driven analysis or search run: a Quick Search, a Deep Research investigation, a Delta comparison, or generating a fix proposal. Investigations draw down your quota, and so does the analysis that runs automatically on an upload. See the investigation unit, below.

Investigation unit

How logcat.ai meters AI usage. Different operations draw down different amounts. Your account gets a budget of units for its period, and your billing page carries the numbers for your account. See Investigation units & quotas.

A single-pass, natural-language lookup that returns a direct answer with cited log lines in a few seconds. Best for triage and known symptoms. See Quick Search.

Deep Research

A multi-step investigation that plans an approach, forms and tests hypotheses, correlates across layers, and produces a citation-backed root-cause report. Best for the hard, open-ended cases. See Deep Research.

Delta

A comparison across separate files (a working capture vs a failing one, or one build against another) to isolate what changed. Distinct from a bundle, which composes related files into one picture. See Delta Comparison.

Source grounding

Resolving a log line back to the place in the platform source that emitted it, so a finding cites a file and line of code as well as a log line. Available for Android and Linux kernel logs when the capture’s platform version matches source we hold. See Source-grounded findings.

Fix proposal

A candidate code change written against the source a finding was grounded to, generated on request. You review and apply it yourself; nothing is ever applied to your code for you. See Candidate fixes.

Subsystem analyzers

Each examines one facet of a log (application, CPU, memory, network, power, security, storage, system services, timeline, system health). Their findings appear on the dashboard, and your questions are answered against them. See Analysis output & dashboards.

How file types are analyzed

  • Text logs (logcat, dmesg, syslog, telecom modem text): ask a question in plain English and get an answer with the log lines cited.
  • Structured captures (pcap, Perfetto traces, ramdumps, SQLite): search queries the records directly, so you can ask for exact counts, sequences, and relationships.

You don’t choose between them. logcat.ai picks the right one from the format, which is why a pcap and a bugreport get different dashboards and different search behavior.

Bundle

A collection of files that belong together (an OEM megadump, a mixed archive, a Qualcomm ramdump) uploaded as one unit. The platform unpacks the archive, analyzes each child, correlates across them, and gives each analyzable child its own dashboard plus a bundle-level roll-up. See Multi-file bundles.

Dashboard

A per-format view of an analyzed file, with charts, an event timeline, and the findings. A file reads as completed once its results are ready to open. See Analysis output & dashboards.

References library

A store of decode artifacts scoped to your account: kernel debug images (a vmlinux or your distribution’s separate debug file), kernel modules (.ko), CAN signal databases (DBC), baseband message databases (.qsr4), and modem firmware images (.elf). Uploading a debug image lets a kernel dump of that exact build be analyzed with symbols; registering a DBC decodes CAN traces against it; a message database restores a modem capture’s compressed narration to text; a firmware image names the function behind trace messages whose text the vendor did not ship. See Kernel dumps & References and Preparing telecom logs.

Insights

Cross-file findings that roll up to the bundle level: patterns no single file’s dashboard would surface on its own, like the same identifier failing across several logs.