Skip to main content
logcat.ai raises $2.55M to build the future of device systems engineeringRead

The pain

Device engineering runs on the slowest loop in software.

Below the application layer, system and OS engineering is still done manually.

01 / 04

The bug nobody can reproduce

A CAN race propagating through VHAL into Android Automotive. A modem PLMN drop only on a specific firmware × carrier combination. A Yocto recipe regression after a meta-layer bump that surfaces three boots in. The fix lives in the relationships between signals. Single-layer tools and grep can't see it.

02 / 04

Cross-layer failures, single-layer tools

Real device bugs cross kernel, HAL, framework, and modem in a single failure. App-layer crash tools stop at the app. Manual log correlation across subsystems is the only option, and it doesn't scale.

03 / 04

The certification gauntlet

Every device fix runs three to six weeks: triage, root cause, patch, regression, CTS/VTS. Per bug. Dozens per release cycle. Most device engineers spend 40–50% of their time here.

04 / 04

Expertise that walks out the door

Cross-layer debugging lives in the heads of two or three senior engineers per team. When they're unavailable, issues queue for days. When they leave, the institutional ability to ship goes with them.

What ships today

Diagnose: autonomous investigation across the device stack.

Upload a bugreport, logcat, dmesg, modem trace, CAN bus capture, or a bundle of them. Get a cited root-cause report back, in minutes.

Investigation engine

Deep Research

Multi-step autonomous investigation across bugreports, logcat, dmesg, modem traces, and CAN bus data. Forms hypotheses, correlates events across subsystems, builds cited reports your team can share.

Ask Why did this happen?

Comparison engine

Delta

Cross-layer multi-file correlation at 1GB+ scale. Compares logs across builds, devices, or time periods to isolate regressions and root causes single-file analysis would miss.

Ask What changed between these two?

Demo · Deep Research investigating a cross-layer issue

Watch Deep Research correlate kernel, framework, and upstream sources, then return a cited root cause.

Under the hood, four pieces compose every investigation: native parsers across every signal format, cross-subsystem correlation, multi-step autonomous reasoning, and line-level citation back to the source.

Read how Diagnose works

Diagnose works today on a real bugreport. The fastest way to evaluate logcat.ai is to bring us one.

See it on a real bugreport

Research preview

Remediate: AI-proposed fixes, human-approved.

Diagnose finds the root cause. Remediate proposes the patch: cross-layer, cited back to the investigation that produced it, gated by the engineer who has to merge it. AI never commits; the human always does. In research preview now for BYOC (bring-your-own-cloud) deployments.

Remediate·RESEARCH PREVIEW
ROOT CAUSE·Watchdog reset on net subsystem after CPU thermal throttle
Proposed patches across the stack
Hand-off · investigation → proposed patch

Outcomes

What this changes for the team that ships.

Engineering-loop outcomes that matter to the people running the release.

Weeks of debugging collapse to minutes

Cross-layer investigations that used to span sprints close in a single sitting. The same artifact, the same root cause, an order of magnitude faster.

Cross-layer investigations any engineer can run

Senior-engineer judgment encoded in the platform, and the investigations live there, not in two or three people's heads. Any engineer can run what previously required the team's deepest cross-layer expertise, and the institutional ability to ship doesn't walk out the door when those engineers do.

Investigations you can hand off — to a vendor, a regulator, a partner OEM

Every finding cites the exact log line, dmesg session, or bugreport section it came from. Failed hypotheses surface as failed, not hidden. Outputs hold up when a chipset vendor, an automotive homologation body, or a customer asks how the conclusion was reached.

Architecture

Built to be auditable.

Citation discipline, hypothesis verification, mandatory human approval, no training on your data. Architectural, not configurable, so the output holds up when a regulator, a vendor, or a partner OEM asks how the conclusion was reached.

Grounding and citation discipline.

Every claim the engine makes cites the source: the exact log line, dmesg session, bugreport section, or device tree node it came from. Findings without citations don't ship. Failed hypotheses surface as failed, not hidden. So reviewers know what was ruled out and why.

Hypothesis verification.

The engine forms hypotheses, gathers evidence, and self-corrects when the evidence pushes back. A hypothesis that fails verification is reported as failed. Not hidden, not silently dropped. The audit trail shows what was tried and why each candidate was kept or rejected.

Mandatory human-approval gates.

An architectural principle, not a feature flag. Diagnose surfaces cited findings; the engineer decides what to do with them. In research preview, Remediate's AI-proposed patches land under the same rule: the model never commits, the human always holds the merge button.

No training on customer data. Ever.

Customer logs are processed in memory for the duration of an investigation and discarded when it completes. They are never used to train, fine-tune, or improve any AI model, neither logcat.ai's nor any provider's. The zero-training rule is contractual with our frontier-model providers via zero-data-retention agreements, and architectural in our pipeline.

Read the full architecture
01
Encrypted at Rest
AES-256 encryption
02
Encrypted in Transit
TLS 1.3
03
Auto-Delete
90-day retention
04
SOC2 In Progress
Type II certification underway
Enterprise · self-hosted

Deployment options

Customer-VPC frontier-model inference means logs never leave your environment. Dedicated cloud instances with custom data retention, SSO/OIDC, and compliance controls. SOC2 Type II certification in progress.

Talk to us

Common questions

Common questions.

The questions device engineering teams ask before pulling logcat.ai into the workflow.

No. Diagnose is autonomous cross-layer investigation: native parsers across every device-software signal format, multi-step reasoning across kernel, framework, modem, and bus, and line-level citation back to the source. Log analysis is the entry point, what an engineer hands the system. The investigation, the correlation, and the cited report are the output.

Diagnose ships today on Delta, our investigation engine: Quick (a fast cited answer), Deep Research (multi-step investigation of one capture), and Delta (multi-file comparison) across bugreports, logcat, dmesg, modem traces, CAN captures, and more. Runtime-to-source attribution, our codebase intelligence layer that grounds investigations in your proprietary source code, is in research preview. Remediate (AI-proposed fixes with mandatory human approval gates) is in research preview for BYOC (bring-your-own-cloud) deployments.

Android: bugreports (zip, 100MB+), logcat in every common format, dmesg, ANR traces, tombstones, dumpsys, event logs. Telecom: QXDM, Shannon, and MediaTek text exports with NAS, RRC, PHY, MAC layer parsing (vendors export binary traces to text via their own tooling; we ingest the text). Automotive: CAN bus traces, VHAL, AAOS logcat. Embedded: dmesg and custom kernel logs across ARM, x86, MIPS, RISC-V. New formats compound the parser corpus. Every customer adds coverage future customers inherit.

Yes. Default deployment is logcat.ai cloud. Enterprise plans include dedicated cloud instances with custom data retention, SSO/OIDC, and compliance controls. Fully self-hosted deployments — BYOC (bring-your-own-cloud) in your own cloud, or on-prem in your datacenter — are available for regulated industries (automotive certification, telecom carriers, defense). SOC2 Type II is in progress.

Custom enterprise contracts with vertical-aware ACVs. An OSS Developer Program exists for open-source maintainers. For commercial deployment, request a technical pilot: we'll bring a real bugreport, run Diagnose on it, and scope the right pilot for your team.

Encrypted at rest (AES-256) and in transit (TLS 1.3). 90-day default retention; configurable on enterprise plans. Log data is never used to train AI models. SSO/OIDC and custom retention policies on enterprise. SOC2 Type II in progress.

Start

Request a technical pilot.

Bring us a real bugreport. We'll show you Diagnose on it. From there, we'll scope a pilot for your team.