# Asama Product -- Your Infrastructure Engineer

Asama is an AI infrastructure engineer for teams who run their own metal. It detects faults across hardware, firmware, OS and virtualization, investigates them to root cause, and executes the fix you approve.

## Detection

Asama watches hardware, BMC, firmware, kernel and virtualization as one system, learning what normal looks like for every host and for its peer group. Four kinds of deviation surface:

### Hardware fault

Hardware degrades before it fails. The warning signs sit in counters nobody checks until the outage. Asama watches every host and compares each to its peers -- for example, SMART pre-failure on a drive whose reallocated sector count has passed the vendor threshold, or NVMe media and data integrity errors that are rising before any read has failed.

### Config drift

Fleets drift quietly. An RMA'd chassis returns with vendor defaults. An emergency change never gets reverted. Each machine looks fine alone, wrong beside its peers. Asama flags settings -- such as BIOS/UEFI PXE options -- that have diverged from the peer cluster, and explains the impact.

### Performance anomaly

The hardest fault to catch isn't a failure or a misconfiguration. The machine just stops performing to spec, with every number in range. Asama compares behaviour such as CPU context-switch rate against the peer cluster and surfaces the host that has diverged.

### Firmware and OS vulnerability

Advisories tell you a version is affected. They do not tell you which of your machines are running it, which are exposed, or which you can safely take down to patch. Asama maps published advisories (for example, vendor BIOS security updates) to the specific servers in your fleet that are affected.

Thresholds only catch what someone thought to set a number for. And the failures that matter cross layers: the symptom surfaces in virtualization, the cause sits in firmware, and the two are owned by different people with different tools.

## Investigate incidents

Ask Asama why something is wrong and it works the problem like an engineer. It keeps state across every layer, so a window that has already closed can be reopened. It tests each candidate cause against the dependency graph and the timeline, and discards what the evidence rules out.

Example: asked "reporting-vm seems slow. Why?", Asama traced the VM's NFS-backed disk to a RAID1 array on another host that was running degraded while an active discard job issued operations against the same device. The cause was never on the server reporting the problem. Each conclusion is backed by evidence: array state, active processes, and VM process load.

Transient issues are difficult to debug. A server drops a burst of inbound packets, recovers on its own, and by the time anyone logs in the evidence is gone. Everything left on that host correlates (the drops, the kernel pressure, the socket errors), but none of it is the cause. The cause is often not on the server reporting the problem at all.

## Remediation workflow

Asama plans the fix against your maintenance policy and sends it for approval. A plan is sequenced into phases (prepare, execute), shows the exact config changes and commands, and defines what happens on success and on failure -- including restoring the original configuration. You can edit the plan, dry-run it, or hand it to the Maintenance Orchestrator.

Change what you want, and every edit becomes an input to the next plan, so plans converge on how your team actually works. Once approved, the Workflow Orchestration Coordinator executes the plan across every step and host, verifying as it goes and looping a human in when verification fails.

The hard parts are the ones around the fix: staying inside the maintenance policy, coordinating with the teams whose workloads sit on those hosts, and knowing what to do at 2am when step four does something nobody planned for.

## Get started

Your infrastructure already has the data. Give it an engineer. Bring one recurring problem your current tooling has not solved, and we will show you how Asama detects it, concludes, and acts.

- Book a demo: https://www.asama.ai/get-started
- Talk to us: https://www.asama.ai/talk-to-us
