Only what needs you, worst first
Every open concern on one page, ranked. Open one, read its evidence, and mark it fixed, mute it, accept the risk or call it a false positive.
How monitoring worksDrill keeps an eye on what you run, follows errors into your code and picks up your alerts — all from one console, with checks you can bend to fit your stack.
The Drill admin console on cluster prod-eu: 75 open concerns, 2 firing alerts, 8 jobs and 506 workloads, with a list of what needs attention and the most urgent jobs.
A read-only kubeconfig and the agent already running in that cluster are all it takes. Drill finds every workload, recognises what it runs, and assesses whatever you point it at.
Checks run on their own, alerts arrive already investigated, and only the decisions that need a person reach you.
Every open concern on one page, ranked. Open one, read its evidence, and mark it fixed, mute it, accept the risk or call it a false positive.
How monitoring worksAdd Drill as a receiver — your existing routes keep paging as before. Turn on automatic investigation and each new alert arrives with its cause, rate-limited so an alert storm can't run away.
Alerts with contextHourly, every weekday at six, or right now. Each run lands in the job's history with what it found and what cleared since last time.
Pick the GitHub or Bitbucket repositories and branches that matter. Drill keeps them indexed, so its answers point at real code — and a fix comes back as a diff, never applied.
How code awareness worksAdd a technology Drill doesn't ship, or change how one is recognised. A rescan re-labels every workload, and the ones you set by hand stay as you set them.
Describe a check the way you'd brief a teammate and it sits beside the built-in ones. Customise any check or playbook for your organization, and reset to the template any time.
The check catalogueWho used Drill, what they asked, what changed and whether every agent is reachable — for every organization, in one place.
Investigations, active users and spend — with trends, models and top users for any range.
Who asked, which model, how many tool calls — and which runs hit tool errors.
Member, sharing, credential, cluster and monitoring changes in every organization, plus impersonation.
Each agent is checked every few minutes; one that can't be reached says why.
Every account across organizations, and read-only impersonation for platform admins.
The whole product in one list — what each feature does, and the options and limits that come with it.
Describe the symptom; Drill queries Kubernetes, logs, metrics, traces, databases and your code to find the cause.
The investigation's steps tick off as it works, with the step in progress marked.
Each query and command is listed as it runs; open any one to read its raw output.
Stop a run at any point and resume it when you're ready.
A short answer, then the evidence, the fix and how to verify it. Queries come copy-ready and diagrams render in place.
Choose the answer style per question; organizations set the default.
Next questions under every answer, one click to ask.
A sensitive command pauses the run and shows exactly what would run. Approve it, or deny it with a note that steers the investigation.
Investigations run in a background worker, not your browser. A closed tab doesn't stop them, and a dropped connection resumes from the results already saved.
The model list comes from your agent and its first entry is the default — no model is built in.
A command palette (⌘K) and shortcuts across the app, each one listed under ?.
Mark an investigation resolved and Drill drafts the symptoms, root cause, fix, verification and failure spread in the background.
Edit any section of the draft, then approve it into the library.
When an answer leans on a past resolution, it names it.
Search resolutions by symptom, error message or service.
Pick the tool calls that cracked a case; IDs and dates in them become inputs.
Describe a method and Drill drafts the skill for you.
Keep a skill to yourself, share it with your organization, or have it applied to every investigation. Run one with /its-name.
For your organization or for any Drill user; they expire on their own and can be revoked any time.
Values that look sensitive are flagged before a link is made; receivers preview what they import.
Choose the repositories and branches that matter; each branch is kept indexed and re-synced on demand.
Answers find a symbol, who calls it and what a change would touch — scoped to your organization's code only.
A self-contained, read-only kubeconfig for discovery, and the agent that runs inside that cluster for the assessments.
Every Deployment and StatefulSet, with the technology it runs recognised from its image and containers. Correct any of them by hand.
Add an engine Drill doesn't ship or tune how a shipped one is recognised; a rescan relabels everything except what you set by hand.
Security posture (Pod Security Standards, RBAC, network isolation, image pinning) or performance & reliability.
One cheap pass over every selected workload's spec, or a full investigation of each one against its technology's playbook.
Workloads picked by namespace, or the cluster itself: control plane, etcd, nodes, scheduling, DNS, storage and capacity.
Leave checks out of one job or re-rate their severity for it — the catalogue itself stays as it is.
Hourly, every 6 hours, daily, weekdays, weekly or any UTC cron — or run it now. Each job picks its model.
30 built-in checks, each citing its source. Add checks of your own; editing one gives your organization its own copy.
PostgreSQL, MySQL, MongoDB, ClickHouse, RabbitMQ, Node.js and Kubernetes — editable, and reset to the template any time.
One problem is one concern across runs: it reopens if it comes back, closes when it's gone, and shows when its severity changed.
Mark it fixed, mute it for 30 days, accept the risk or record a false positive — with a note where it matters.
Every run with what was new, resolved and still open, its tool calls, its duration and its cost.
Open concerns, firing alerts, jobs and workloads at the top; what needs attention, worst first; overdue jobs flagged.
Drill sits beside your existing routes, which keep paging as before.
Each alert lands on the workload it's about, next to the concerns Drill had already raised there.
Opt-in per cluster, and rate-limited so an alert storm can't run away.
Start one from any alert's page whenever you choose.
Investigations, active users and spend — with trends, models and top users, per organization or across all of them.
Every investigation: who asked, which model, its tool calls, and whether any tool failed.
Member, sharing, credential, cluster and monitoring changes in every organization, plus impersonation.
Every agent checked every few minutes, with the reason when one can't be reached.
Members, agents, clusters and spend per organization; runs, spend and last activity per user.
Platform admins see Drill exactly as a user does, behind a banner, without changing anything.
Each team's skills, resolutions, agents, clusters and code stay in its organization; investigations are private to whoever ran them.
Owner, Admin and Member, invited with single-use links; a role change applies on the next request.
Installed with Helm. No telemetry, no phone-home.
The agent's Kubernetes role is get, list and watch, with no access to Secrets. Fixes come back as diffs for you to apply.
OpenAI-compatible endpoints including local servers, Azure AI Foundry, AWS Bedrock in your region, or Vertex AI.
Agent keys, git tokens and kubeconfigs are encrypted at rest; webhook, invite and share tokens are stored only as hashes.