For technical support

Close complex tickets
without escalating

Runtime gives support an agent with its own computer. It runs queries, writes scripts, and reproduces the incident, then waits for your team to approve the fix.

Book a demo

Start for free

Runtime session · support-agent

Priya Nair

acme.com's webhook deliveries have been failing since Tuesday, and now they say they were double-charged. Ticket #8812.

Support agentRuntimeworking

On it. Spinning up a computer with the billing repo.

Read Zendesk ticket #8812Deliveries failing since Tuesday

Correlated webhook delivery logs42 failures, all POST /webhooks

Queried billing.chargesAcme charged twice, 84s apart

Read billing/worker.tsRetry path skips the idempotency key

Wrote repro.ts and ran itReproduced the double charge

5 tool calls · bug reproduced

Reproduced it. The retry path in billing/worker.ts re-processes a webhook without the idempotency key, so Acme was charged twice. I wrote the patch and drafted the $1,188 refund, plus refunds for the 11 other accounts hit by the same bug.

Approve refundsOpen PR with the fix

Message support-agent

Supporting customers at fast-growing technical companies

Like a new engineer on the team, without the onboarding

Ask it anything you would ask an engineer, in plain English. It already has your repos, your data, and your tools, so it can go and check.

Tickets close where they land

The agent does the work while your analyst stays with the customer. No handoff, no waiting on another team to pick it up.

Escalations become the exception

Engineering hears about real bugs, with the root cause already written up. Not lookups, log pulls, and one-off queries.

Every reply is backed by evidence

Customers get answers with the query, the log line, and the code behind them. Support reviews the trail before it ships.

Powerful underneath, safe by default

It does the work, not just lookups

It gets its own computer, so it runs queries against your data, writes throwaway scripts, and reproduces the incident to prove the cause.

GitHub

Stripe

Datadog

Snowflakeand more

support-agent · ticket #8812working

Ranstripe events list --delivery-failed

Queriedselect * from billing.charges where org=...

Readbilling/worker.ts, found the retry loop

Wroterepro.ts, reproduced the double charge

Draftingroot cause + customer reply

It takes action, on your terms

It can issue the refund, apply the fix, or update the record. Anything you mark sensitive stops and waits for a named human first.

Approval gatesScoped write accessEvery action logged

Waiting on you

Refund $1,190 to Acme Corp

duplicate charge, retry bug in billing/worker.ts

Approve refundReject

Customer reply draftedReview

Root cause filed for engineering1 file changed

Plugged into your support stack

Tickets arrive through the helpdesk you already run. The agent investigates with the tools your engineers already use.

Where tickets land

Your helpdesk stays the queue

The agent picks the ticket up where it already is, works the thread, and posts the answer back in the same place.

Zendesk

Jira

Intercom

Pylon

Slack

HubSpot

Salesforce

What it investigates with

The same tools an engineer opens

Read the code, pull the charge, correlate the logs, query the warehouse. Anything with an API can become a tool.

GitHub

Stripe

Datadog

Sentry

Snowflake

BigQuery

Postgres

Linear

From ticket to resolution

Connect your helpdesk and one repo, set the permissions once, and run your first ticket the same day.

A complex ticket lands

In Zendesk, Jira, Intercom, Slack, or email. Support tags the agent instead of escalating.

The agent reproduces it

On its own computer. It runs the queries, writes a script, and confirms the cause.

It proposes the actions

The refund, the fix, and the customer reply, each with the evidence behind it.

Support approves, it ships

One click. The action runs, the customer gets the answer, engineering gets the write-up.

See it on one of your tickets

Safe to give a support team

Read about our security

AuditedSOC 2Compliant

Read-only until you say otherwise

The agent gets the scopes you assign and nothing else. It can read the charge and the log line without the ability to change either.

Approval gates on sensitive actions

Refunds, account changes, and writes to production pause for a named human. Support sees exactly what it is approving.

Every investigation is on the record

Each run is stored with its steps, queries, sources, and cost, so a lead can review how any ticket got closed.

Frequently asked questions

01Does support get access to production systems?

No. The agent works in a sandbox with the scopes you set. Reads are on by default, and writes or refunds pause for a human approval.

02What can the agent actually investigate?

Anything you connect: repos, logs, billing, and your warehouse. It reads code, runs queries, and correlates them the way an engineer would.

03How do we know the answer is right?

Every reply shows the steps, queries, and sources behind it. Support reviews the trail before anything reaches the customer.

04Does this replace our helpdesk?

No. Runtime plugs into Zendesk, Intercom, or whatever you run today. The agent works the tickets, your helpdesk stays the queue.

05What happens when the root cause is a real bug?

The agent writes up the root cause with the evidence and can draft a fix for engineering to review. The escalation arrives already solved.

06How fast can we start?

Connect your helpdesk and a repo, set the permissions once, and run your first ticket the same day.

Close the next complex ticket without an escalation

Give your support team the same tools your engineers use, with the permissions you set. No credit card required.