AI support engineer · senior-level RCAPrivate beta

The bug with no error and no stack trace just landed. Again.

Treap is the AI support engineer that takes the escalation and investigates it like your best senior would, then answers in Slack with the root cause and the evidence.

See how it works ↓

Tier-1 bots deflect the ticket.
Treap investigates it like a senior engineer.

// what it posts back

# eng-escalations
Riya (Support) 10:42

Acme says their March invoice total is $40 too high. No error on our side, nothing in Sentry. Known bug?

Treap APP 10:43

Investigating…

root cause   Tax is rounded per line item, then summed, double-counting half-cents. Introduced in #4127.
likely PR     #4127 "Refactor invoice tax calc", deployed Mar 2, 14:10
evidence    
  • Acme invoices skew high only after deploy 8f3c1a2
  • roundLineItem() called before sum(), order flipped in #4127
  • 17 of 92 March invoices affected, all multi-line
confidence0%
▶ see it live on a 15-min call

The problem

"It's broken," and there is nothing to grep for.

01

No error. No event.

A customer says the invoice total is wrong, or their data disappeared. Nothing threw. Nothing logged. There is no exception for your monitoring to catch, just a person telling you something is off.

02

A senior engineer stops everything.

They spelunk through GitHub history, log lines, and the production database to reconstruct what actually happened. The most expensive person on the team, doing manual archaeology.

1 to 3 hrs of focused senior time, per incident
03

"Is this a known bug?"

PMs and support keep interrupting mid-dig. Half the time it is config drift or a regression you already fixed once, but nobody remembers, so you find it from scratch again.

How it works

From a Slack message to a root cause.

1

Post the escalation in Slack

Paste the customer report into the channel Treap watches, or @-mention it on the thread where support already flagged it. No ticket format, no error required.

2

Treap investigates, read-only

It correlates across GitHub, logs, your database, and Sentry, reading only and never writing, to reconstruct what changed and when, the way your senior engineer would.

3

Replies in-thread with a verdict

Suspected root cause, the likely PR or deploy, the exact evidence it checked, and a confidence score. Trust it, or take over knowing exactly where it already looked.

4

Saved to incident memory

Every investigation is remembered. When a recurring issue returns, Treap recognizes it: "this matches the rounding regression from PR #4127, fixed and reintroduced."

// what it posts back

# eng-escalations
Riya (Support) 10:42

Acme says their March invoice total is $40 too high. No error on our side, nothing in Sentry. Known bug?

Treap APP 10:43

Investigating…

root cause   Tax is rounded per line item, then summed, double-counting half-cents. Introduced in #4127.
likely PR     #4127 "Refactor invoice tax calc", deployed Mar 2, 14:10
evidence    
  • Acme invoices skew high only after deploy 8f3c1a2
  • roundLineItem() called before sum(), order flipped in #4127
  • 17 of 92 March invoices affected, all multi-line
confidence0%

Not a dashboard. A teammate that sits in front of your engineers.

Treap is your first line of technical support, at senior depth. It takes the first pass at every escalation, so a report only reaches a human when it actually needs one.

Triages every escalation first

Incoming reports get sorted before anyone is pulled in: real bug, known issue, config drift, or expected behavior. The noise stops at Treap.

Reads the code, returns an RCA

It reads the actual code and data, reconstructs what happened, and replies with a root cause you can act on, not a guess.

Answers the rest of the team

Support and product can ask "is this expected?" or "did we ship X?" and get a grounded answer, sparing your engineers the interruption.

It remembers every investigation, and gets faster and cheaper each time

Every investigation is saved to incident memory. Ontology-backed memory plus agentic RAG retrieve only what the next investigation needs, so recurring issues come back in seconds and each run stays token-efficient, not brute-force context stuffing.

matches PR #4127 · rounding regression, seen 2x · token-efficient recall

Connectors

Reads your whole stack. Answers where you already work.

Connect read-only access to the sources you run. Treap correlates across all of them in one investigation and replies in the tools your team already lives in.

Sources it readsread-only
Code
GitHubGitLabBitbucket
Databases
PostgreSQLMySQLMongoDB
Logs & monitoring
OpenSearchSplunkSentry
Treapreads across every source in one investigationread-only · correlated
Where it answers
Chat
SlackMicrosoft TeamsGoogle Chat
Email
Gmail
Tickets & docsread + write
JiraConfluence

Jira & Confluence work both ways — pulled in as evidence during an investigation, then updated with the findings.

On the roadmap
DatadogNew RelicBigQueryLinear

Why Treap

There are great tools. None of them fit you.

vs. Sentry Seer and error-triage AI

It needs an event. Your bug doesn't have one.

Them

Starts from an error or exception Sentry already captured, then triages it.

Treap

Starts from a human message, "this number is wrong," with no event to anchor to.

vs. Cleric, Resolve and enterprise AI SRE

Built for big orgs, sold by a sales team.

Them

Six-figure contracts, a demo call, a pilot, and a platform team to justify it.

Treap

Self-serve, flat $149 a month. Connect read-only access and run an investigation today.

vs. AI support bots and helpdesk AI

They deflect the ticket. They can't investigate it.

Them

Canned answers and help-center search. No access to your code or data, so anything real bounces back to engineering.

Treap

Reads your code, logs, and database and returns a senior-grade root cause, not a deflection.

Treap is Cleric and Resolve for the rest of us. AI technical support that does senior-level RCA for 10 to 50 person SaaS teams: deep enough to find the real cause, self-serve and transparently priced, with no six-figure contract and no procurement.

You are not handing prod to a black box.

Read-only by design

Treap connects with read-only, scoped credentials you issue. Its toolset has no write, deploy, or mutate actions, and every action it takes is a read, logged in full.

Bring your own key

Point Treap at your own LLM key. Your code and data are used to investigate, never to train a shared model.

Self-host option

Run it in your own VPC if you would rather nothing leaves your network. Scoped credentials, full audit log of every read.

FAQ

The questions engineers actually ask.

What if it's wrong?

Treap never gives a bare answer. Every reply ships with the exact evidence it checked and a confidence score, so you can trust it or take over knowing precisely where it already looked. It reads only, so a wrong guess can never change anything.

Is my code or data used to train a shared model?

No. On bring-your-own-key, your code and data go to your own LLM key and nowhere else. Treap connects read-only and uses your data to investigate, never to train a shared model.

Which stacks and databases does it support?

Code hosts (GitHub, GitLab, Bitbucket), databases (Postgres, MySQL, MongoDB), and log and monitoring stores (OpenSearch, Splunk, Sentry) today, with Datadog, New Relic, BigQuery, and Linear on the roadmap. You connect a source with a read-only credential.

How long does setup take?

Minutes. There is no agent to deploy and no sidecar. Drop in read-only credentials and Treap investigates across them. In private beta we get you running on a short onboarding call.

BYOK or managed, which should I pick?

Bring your own key is a flat $149 a month per team if you already run an LLM provider. Managed is usage-based, billed per 1M tokens, if you would rather not manage a key. Both are self-serve and read-only.

Can we self-host?

Yes. Run Treap in your own VPC if you would rather nothing leaves your network. Scoped read-only credentials, full audit log of every read.

Pricing

Two ways to run it. Both self-serve.

One senior investigation costs you 1 to 3 hours of your best engineer. One investigation pays for the month.

Managed
Usage-basedbilled per 1M tokens

We run the model. Pay only for what you investigate.

  • No LLM key to manage
  • Token-efficient by design, so each run stays cheap
  • Read-only connectors and incident memory

Metered per 1M tokens. No minimum.

Stop doing manual archaeology.

Request a spot in the private beta. We onboard a few teams a week and get you running on a short call, and design partners get the first real investigations.