Customer-resolution agent

Closed isn’t
fixed.

Rectify carries a customer issue across Gmail, GitHub and Slack, runs the customer’s own workflow to check it really works, and follows through until recovery is observed.

See how a case closes
Fragmented records passing through a verification engine and emerging complete
01 · Recovery engineComplaint to observed success
An empty fractured export transformed into a verified export with complete rows
StatusObserved proof
Northstar Research · CSV exportDemo scenario
  • GitHubEngineering issue
    Closed
  • SlackRollout message
    “Rollout complete”
  • ReportDeskExport response
    HTTP 200
  • Rectify checkCustomer workflow
    Still failing · expected records missing

Every system says it’s fixed. The customer still can’t export. Rectify runs the customer’s workflow instead of trusting a status.

Four facts, never merged

Most tools close a customer issue when a ticket closes. Rectify keeps each fact separate and only moves forward on evidence.

01

Engineering finished

What GitHub and Slack claim. Recorded as a statement, never as proof.

02

Workflow verified

Rectify runs the customer’s real export for their tenant and checks the exact expected records.

03

Customer notified

An approved message is sent. Sending is not the same as the customer succeeding.

04

Recovery observed

Only the product server seeing the customer complete the task closes the loop.

One case, three apps

Rectify works where your team already does. Each app gets a narrow set of actions, checked on the server before anything is written.

Email, engineering, and conversation evidence converging into one verified case
02 · Evidence constellationOne case across three apps
Gmail

Complaint to reply

Opens the case from the customer’s thread, drafts the reply, and sends only the exact message a human approved.

GitHub

Evidence to handoff

Matches the engineering issue, then opens one customer-impact issue with expected versus observed results.

Slack

Handoff to approval

Tells engineering what is still failing and asks an approver to sign off on the exact customer message.

The model proposes. Policy decides.

The agent searches, compares evidence and drafts. Server-side policy authorizes every action, and observed evidence is the only thing that changes a case.

A customer message passing through a single protected approval path
03 · Approval bindingOne exact message
Approval

Consent binds the exact message

Recipient, subject, body and the passing check are sealed into the approval. Any change needs a new one.

Identity

Customers come from a trusted directory

Names and signatures in an email can’t pick the tenant or add a recipient. Ambiguity goes to a human.

Writes

Intent is saved before every write

If a send’s response is lost, Rectify checks the provider before acting and never blindly retries.

Content

App content is untrusted input

Instructions hidden in emails or issue comments can’t bypass approval or widen what the agent can touch.

Built at the Multi-App AI Agent Hackathon

Comma Capital