Product

A probe that sits in the error, not in a survey you send later.

AethelLayer watches your CLI, Web SDK, and API. When a developer hits a real failure, we keep the stack, ask one question, and draft the ticket. You still hit send. Memory Graph and Intelligence unlock on Growth, Scale, and Enterprise.

CLI
$ npx @acme/cli login
ERR_DOCKER_OOM
probe: what were you doing?
SDK
probe.wrap(client)
JWT expired at +46s
callback never returned
API
POST /oauth/callback
gRPC deadline exceeded
draft #482 waiting

CLI probe

Wrap the command they already run. When compose dies or auth login OOMs, the probe already holds the command, the OS, the exit code, and the compose memory cap — not a screenshot they forgot to attach.

Web SDK probe

One wrap around your Node, Python, or browser client. JWT expiry, OAuth popups, and install friction land as threads, not as a red bar in a dashboard they will never open.

API middleware

Hang it on Express, FastAPI, or Next.js instrumentation. gRPC deadlines and 429s become a conversation while the caller is still in the request. Read-only to your product.

What ships on every plan

Catch. Ask. Draft. Approve.

Starter is not a teaser of the inbox. It is the full capture loop: probes on three surfaces, a developer thread, GitHub drafts you still approve, and analytics on what failed. What Starter does not include is Memory Graph — the constellation of clusters, playbooks, and Graph-RAG context that starts on Growth.

Probes

CLI, Web SDK, API middleware. Bearer API key. Monthly quota by plan.

Inbox

One thread per error event. Filters for CLI, OAuth, rate limit, resolved.

Drafts

GitHub and Linear markdown with taxonomy, stack, suggested fix. You hit send.

Watch

Pattern rules: JWT spikes, Compose OOM twice in a session, 429 before noon.

Inbox

Intercom, except the people in it just hit your SDK.

Each thread is one failure. You are not guessing from a funnel. The right rail holds environment specs and, on Growth+, graph correlations plus session replay.

  • One failure, one thread. Stack, runtime, OS, endpoint, taxonomy, and the sentence they typed. If they walk away, you still have the capture. If they answer, you have intent.
  • Hashed identity. We cluster developers, we do not dossier them. No source dumps. Org members see only their workspace.
  • Replay next to the thread. One-click PostHog and Mixpanel deep links attach from developer + error identity — Growth and above — so GTM and DX share the same moment.
  • Live, not batched. The inbox updates as the probe talks. You are not waiting for a nightly export to find out an evaluator left at noon.

ERR_DOCKER_OOM

auth-worker

JWT_EXPIRED

auth-worker

gRPC deadline

auth-worker

live · 212ms

Caught OOM on /oauth/callback

CLI login inside Compose for a staging cut

Draft GitHub #482 · waiting on you

Tickets

The issue writes itself. You still own send.

GitHub

A markdown draft in the language your eng team already uses. Taxonomy, stack, suggested fix. Agents propose. Humans ship.

Linear

Same payload as a ticket your roadmap already understands. On Growth+, clusters can become draft epics you push yourself.

Approve first

Nothing auto-files. If a product writes issues on its own, it is not a teammate. It is a mess.

Memory Graph · Growth+

The next probe is not a cold start.

Memory Graph is Graph-RAG over errors, runtimes, solutions, and developers. Nodes and edges remember that ERR_DOCKER_OOM correlates with auth-worker, Node 20, and a 512Mi Compose cap — and that raising container memory resolved it for fourteen unique developers. Starter workspaces see a Growth lock. Growth, Scale, and Enterprise open the atlas.

Clusters, not incidents

Same error code across CLI, Web, and API collapses into one constellation. You fix a class of failure, not a single stack.

Context on the thread

Graph pills on the inbox rail: CORRELATED_WITH, CAUSED_BY, RESOLVED_BY — with confidence, not vibes.

Ingest from probes

Each capture can write nodes and edges. Playbooks match against the graph so known clusters reply without a human.

See it in the dashboard after you upgrade, or compare plans. Starter does not include Memory Graph.

Intelligence · Growth+

From a thread to an operating system for DX.

Intelligence sits on top of probes and the graph. It is how a DX org runs Monday: friction by surface, playbooks armed, eval keys for sales, a churn digest, a public health badge, competitor-win alerts, roadmap epics, bots, and onboarding drop-off.

Friction score

Weekly CLI vs Web vs API risk index from probe volume, unresolved ratio, unique developers, and time-to-resolve.

Auto-reply playbooks

Graph-matched replies that ship without a human for known clusters — Docker OOM, JWT skew, starter 429s — with a confidence gate.

Eval-mode keys

Temporary elevated probe quotas for sales PoCs without opening Growth for the whole org.

Why they churned

Monday digest: top unresolved clusters plus lost evaluators, queued as email when you want it sent.

SDK health badge

Public status snippet — “OAuth path: healthy / degraded” — embeddable SVG for docs.

Competitor-win alerts

Same error code plus “installed X then left” for GTM. Ack or dismiss in Intelligence.

Taxonomy → roadmap

Auto-group Linear epics from Memory Graph clusters. You still push the epic.

DX bots + heatmaps

!probe ERR_DOCKER_OOM returns graph context and a draft fix in Slack or Discord. Onboarding heatmaps show which quickstart step loses unique developers.

What we are not

Keep Datadog. Keep Mixpanel. Keep GitHub.

Not observability

We do not replace traces, metrics, or error budgets. We attach intent to the failure those tools already know died.

Not a chatbot

The developer sees one question in the terminal or SDK error UI — not a landing page named “assistant.”

Not auto-filing

Drafts wait. Playbooks can auto-reply in-thread. They do not open public issues without you.

See it fail on purpose

The walkthrough runs a staging login, catches ERR_DOCKER_OOM, and shows the draft. Ten minutes if you want it in your own CLI. Memory Graph is the Growth+ step after that.

Early access

Stop finding out about churn in a dashboard three days later.

No self-serve signup. Book a Calendly slot, or send details to sales.