Skip to content

For engineering, SRE and IT operations leaders

Shorter incidents, with a person approving every fix

Your incident commander asks Tiersel what is broken and gets the failing service, the deploy that preceded it, the runbook and a status page draft. On-call engineers get the night's alerts grouped into a few real issues. Rollbacks, silences, access grants and postmortems all wait for a named person to approve them.

Live demo on synthetic data · Built on Jev by TypeSafe AI

Preview

An animated example types requests as if they were spoken and shows the card Tiersel builds for each one.

For the VP of Engineering, Head of SRE, CIO or IT service management lead

The first twenty minutes of every incident go to finding things.

When a page fires, the responder opens the APM tool, the deploy history, the logs, the runbook wiki and the incident channel, and stitches them together by hand while customers wait. Overnight, one real problem can arrive as dozens of alerts, and the noisy ones train people to ignore pages. Production access requests queue behind a busy approver or get granted too broadly. The postmortem gets written days later from a scrolled-back chat log. Your systems already hold every fact needed. Nobody has time to assemble them while the incident is still open.

4 agent workflows

Built for the work your team already does.

Each workflow starts from a sentence. The agent calls your systems, shows its evidence, and waits for a person before anything consequential happens. Every card below is the real, working interface.

Workflow 01 · Incident commander or primary on-call

Incident command during an outage

Incident commandIT operations
  1. 01 · Trigger

    A service starts failing

    An SLO burn alert pages on-call, or the incident commander asks out loud what is going on with checkout.

  2. 02 · Tool call

    Read errors and traces

    Pulls error rate and latency from APM, finds the failing spans and the downstream calls they time out on.

  3. 03 · Tool call

    Check what changed

    Reads deploy history and finds the release that reached the service 14 minutes before errors began, with its diff.

  4. 04 · Reasoning

    Name the likely cause

    Correlates error onset with the new build, maps impacted services and fetches the rollback runbook, with a confidence score and the evidence.

  5. 05 · Human approval

    Commander approves the rollback

    The rollback to the last good version and a status page draft are prepared. Nothing changes in production until the commander approves.

    Nothing happens until a person approves
  6. 06 · Outcome

    Mitigated with a record

    The rollback runs, the runbook steps are ticked off, and every action lands on the incident timeline for the postmortem.

What changes

The commander starts from the failing service, the suspect deploy and the runbook instead of six browser tabs, and every production change carries an approval.

APM and tracingDeploy pipelinePaging and on-callRunbook wikiStatus page

Workflow 02 · On-call engineer or SRE lead

On-call handoff and alert tuning

On-call summaryIT operations
  1. 01 · Trigger

    The shift ends

    The outgoing or incoming on-call engineer asks for a summary of the night.

  2. 02 · Tool call

    Collect alerts and pages

    Reads every alert event and page in the window from the monitoring and paging tools, plus the on-call schedule.

  3. 03 · Reasoning

    Group and find the noise

    Groups alerts by fingerprint and service, computes MTTA and MTTR, and flags monitors that flapped without anyone acting on them.

  4. 04 · Human approval

    Engineer approves the silence

    A 24-hour silence for the flapping monitors is prepared with a tuning suggestion. It applies only when the engineer approves.

    Nothing happens until a person approves
  5. 05 · Outcome

    A clean handoff

    The day shift starts with a short list of real issues and open follow-ups, and the noisy monitors are queued for tuning.

What changes

A night of raw alerts becomes a handful of grouped issues with owners, and monitor tuning starts from evidence instead of complaints.

Monitoring and alertingPaging and on-callTicketing

Workflow 03 · Engineering manager, SRE or security approver

Just-in-time production access

Access requestIT operations
  1. 01 · Trigger

    An engineer needs production access

    A request arrives in chat or by voice for read access to a production database during an investigation.

  2. 02 · Tool call

    Check identity and ticket

    Looks up the requester, their team and manager in the identity provider, and confirms a linked ticket is open and assigned to them.

  3. 03 · Tool call

    Evaluate policy

    Checks least privilege, the time limit, manager approval and recent strong authentication, and finds PII columns to mask.

  4. 04 · Reasoning

    Score the risk and narrow scope

    Scores the request, recommends the narrowest role that fits the ticket, and shows the exact grant as a diff.

  5. 05 · Human approval

    Approver grants it

    A time-boxed grant is prepared with an automatic expiry. Access is created only when the approver says yes.

    Nothing happens until a person approves
  6. 06 · Outcome

    Access that expires on its own

    The grant revokes itself at the stated time, and the request, policy result and approval are logged to the ticket.

What changes

Production access is granted in minutes at the narrowest scope that fits the ticket, and nothing stays open past its window.

Identity providerCloud IAMDatabase access proxyITSM

Workflow 04 · Incident owner or engineering manager

Postmortem draft from the incident record

Postmortem draftIT operations
  1. 01 · Trigger

    The incident is resolved

    The incident owner asks for the postmortem draft the day after.

  2. 02 · Tool call

    Rebuild the timeline

    Reads the incident record, the incident channel, deploy diffs and service metrics to build a timestamped timeline.

  3. 03 · Tool call

    Measure the impact

    Counts failed requests and affected customers and reads how much error budget the incident used.

  4. 04 · Reasoning

    Draft cause and actions

    Writes the summary, root cause and contributing factors in blameless language, and proposes action items with owners and due dates.

  5. 05 · Human approval

    Owner reviews and publishes

    The owner edits the draft, picks which action items become tickets, and approves publishing.

    Nothing happens until a person approves
  6. 06 · Outcome

    Actions tracked, not forgotten

    The review is posted to the incident review channel and each approved action item becomes a ticket with an owner.

What changes

The postmortem is drafted while the details are fresh, and its action items land in the backlog with owners instead of in a document nobody reopens.

Incident managementChatAPMDeploy pipelineTicketing

How it plugs in

An agent on your systems, one component per workflow.

Requests come in by voice, text or system event. The router picks the workflow, the agent calls your systems, and the answer lands as a live card where your team works. Nothing is written back until a person approves it.

In

VoicePush to talk, any browser
Typed requestSame intent, same card
System eventWebhook, queue or schedule

Tiersel

Intent routerPicks 1 of 4 workflows as you speak
Incident command during an outageIncident commander or primary on-call
On-call handoff and alert tuningOn-call engineer or SRE lead
Just-in-time production accessEngineering manager, SRE or security approver
Postmortem draft from the incident recordIncident owner or engineering manager

Your systems

Observability and APMDatadog, New Relic
Paging and incident managementPagerDuty, Opsgenie
CI/CD and deploysGitHub Actions, Argo CD
ITSM and ticketingServiceNow, Jira Service Management
Identity and accessOkta, Microsoft Entra ID

Out

Live cardIn your app, sidebar or kiosk
Human approvalCommander approves the rollback
Write-back + auditEvery call and decision logged

Workflow blueprint · Incident commander or primary on-call

Incident command during an outage

Starts when
A service starts failing. An SLO burn alert pages on-call, or the incident commander asks out loud what is going on with checkout.
Reads
Read errors and traces. Pulls error rate and latency from APM, finds the failing spans and the downstream calls they time out on.
Check what changed. Reads deploy history and finds the release that reached the service 14 minutes before errors began, with its diff.
Decides
Name the likely cause. Correlates error onset with the new build, maps impacted services and fetches the rollback runbook, with a confidence score and the evidence.
Waits for
Commander approves the rollback. The rollback to the last good version and a status page draft are prepared. Nothing changes in production until the commander approves.
Writes back
Mitigated with a record. The rollback runs, the runbook steps are ticked off, and every action lands on the incident timeline for the postmortem.
APM and tracingDeploy pipelinePaging and on-callRunbook wikiStatus page

This workflow is one component

export const incident = defineSkill({
id: "incident", category: "ops",
// when this workflow, and not a neighbour, is meant
criterion: "An engineer or incident commander asks about a live production incident or outage: what is failing, the likely cause, impacted services, the runbook, a status page update or a rollback. Not a past incident's postmortem, not overnight alert summaries, and not a home or insurance incident.",
examples: ["what's going on with the checkout outage", "why is the payments api throwing errors"],
match: (text) => score, // intent score, runs on every word
parse: (text, ctx) => record, // entities + tool calls into your systems
reply: (record) => "…",
Card: IncidentCommandCard, // the live UI, with its approval step
});

The id, criterion, examples and reply are read live from the skill that renders the card above. Adding a workflow for your team means adding one of these: its intent, the calls it makes, what it says, and the interface your people approve in.

Fits your stack, answers to your people

Reads your systems. Acts only with approval.

What the agent connects to

Observability and APM

DatadogNew RelicGrafanaDynatrace

Paging and incident management

PagerDutyOpsgenieincident.io

CI/CD and deploys

GitHub ActionsArgo CDSpinnaker

ITSM and ticketing

ServiceNowJira Service ManagementLinear

Identity and access

OktaMicrosoft Entra IDAWS IAMHashiCorp Vault

Chat and status page

SlackMicrosoft TeamsAtlassian Statuspage

Product names are examples of systems in this category. Each deployment is connected to your systems through their APIs as part of the build.

A person approves every production change

Rollbacks, alert silences, access grants and published postmortems are prepared by the agent and carried out only after a named person approves them.

Every step is logged

Each tool call, the evidence it returned and the approval or rejection are written to the incident or ticket, so anyone can replay how a decision was made.

Synthetic data in this demo

Every service, engineer, incident and number on this page is invented, so nothing you try here touches a real system.

Least privilege, in your environment

When deployed, Tiersel acts only through the scoped, read-first credentials you grant, and logs and customer data stay in your environment.

The cost of doing it by hand

Put your own numbers in.

Move the sliders to match your operation. Nothing is sent anywhere unless you choose to email it.

Each year

672

staff hours back, about 0.3 full-time people

$67,200

of staff time, at your hourly cost

Email me a pilot plan for these numbers

Counts only engineer time spent gathering context, posting updates, building the timeline and writing the postmortem, summed across responders. It excludes the cost of downtime itself and assumes no change in incident count. The hourly default starts from the BLS mean wage for software developers, $71.20 in May 2025, plus about 40 percent for benefits and overhead.

Hours = incidents a month × 12 × minutes each × share, ÷ 60. A full-time person is 2,080 hours a year. Default hourly cost from U.S. Bureau of Labor Statistics, OEWS May 2025, software developers, 2026.

Questions buyers ask

Straight answers.

  • No. Tiersel sits on top of the tools you already run. When deployed, it reads from them and writes back through their APIs with the permissions you grant. The demo on this page uses synthetic data and connects to nothing.

  • No. It prepares the rollback, silence, grant or publish step and shows the evidence. Nothing changes until a person approves it, and the approval is logged with the incident or ticket.

  • It lines up error onset with recent deploys, config changes and failing traces, and shows each piece of evidence with a confidence score. It points the commander at the likely cause. The commander still decides.

  • A pilot starts read-only: metrics, deploy history, alerts and the incident record. Write actions such as rollbacks or access grants are added one at a time, each scoped to the smallest role that does the job.

  • When deployed, they stay in your environment, and each tool call asks only for the fields the workflow needs. In this demo there is no real data at all.

  • Try the live demo on this page now. A pilot starts with one team, one service and the on-call summary, read-only, so your engineers can compare it with their own handoff before anything writes back.

See it on your own workflow.

Watch the it operations workflows run end to end, or speak to the agent yourself. Then tell us which one to build on your data.

Scope a pilot with hello@sasid.ai
  1. 01

    Pick one workflow

    A short call to choose the workflow that costs you the most time, and who approves its actions.

  2. 02

    See it on your data

    We build that workflow against your sample or de-identified data so your team judges the real thing.

  3. 03

    Run it with people in charge

    Your staff use it on live work. The agent drafts and prepares; a person approves every consequential step.

  4. 04

    Decide from results

    You keep what measurably saves time and drop what does not.