Architecture

How pmkit Works

pmkit runs multi-step PM workflows with AI agents while keeping humans in control. Draft-only by design, with full traceability and enterprise governance.

Core Principles

Four principles guide how pmkit handles AI automation in enterprise environments.

Draft-Only

Agents never write directly to external systems. Every change is a proposal that requires human approval.

Full Traceability

Every tool call, every source, every artifact is logged. Trace any insight back to its origin.

RBAC & Governance

Role-based access control with permission simulation. Audit logs for compliance.

MCP Architecture

Swap mock connectors for real ones without changing job logic. Standardized tool integration.

Step by Step

From connecting tools to reviewing proposals; here's how a pmkit workflow runs.

01

Connect Your Tools

MCP (Model Context Protocol) connectors link pmkit to your existing tools; Jira, Confluence, Slack, Gong, Zendesk, and more. Each connector authenticates securely and logs every interaction.

  • OAuth 2.0 or API key authentication
  • Encrypted credential storage
  • Per-tool permission scoping
  • Audit logging for every API call
Integrations
Illustrative

Jira

Demo (Mock)

Confluence

Demo (Mock)

Slack

Demo (Mock)

Gong

Demo (Mock)

Zendesk

Demo (Mock)

Amplitude

Demo (Mock)

In demo mode, all connectors return simulated data. Contact sales to connect real tools.

02

Configure Jobs

Set up the PM jobs you want to run; daily briefs, meeting prep, VoC clustering, competitor intel, roadmap alignment, or PRD drafts. Configure sources, schedules, and output destinations.

  • Six pre-built job types
  • Custom job templates (Enterprise)
  • Schedule-based or on-demand execution
  • Source and output configuration
Job Configuration

Daily Brief

Every day at 8am

Active

VoC Clustering

Every Monday

Active

Competitor Intel

Every Friday

Active
03

Run with Traceability

When a job runs, pmkit executes a multi-step workflow; calling tools, synthesizing data, and generating artifacts. Every step is logged with timing, inputs, and outputs.

  • Real-time job progress tracking
  • Tool call timeline with durations
  • Source citations in every artifact
  • Error handling with graceful fallbacks
Job Timeline
slack.get_messages142ms
jira.get_sprint89ms
zendesk.get_tickets156ms
gong.get_insights201ms
04

Review Proposals

All external writes are proposals, not direct actions. Review the diff, edit if needed, then approve to publish to Jira, Confluence, or Slack.

  • Preview before publishing
  • Inline editing capabilities
  • Approval workflow with RBAC
  • Proposal expiration policies
Pending Proposals
Create Jira Epic
Pending

Search Filters Feature • 3 stories

Post to Slack
Approved

Daily Brief • #product-team

Six Cadence Jobs

Pre-built workflows for the most common PM tasks. Run on-demand or on a schedule.

Daily Brief

Daily

Synthesize overnight activity from Slack, Jira, support, and community.

Meeting Prep

On-demand

Prepare for customer meetings with context from calls, tickets, and CRM.

VoC Clustering

Weekly

Cluster feedback from support, calls, and community into themes.

Competitor Intel

Weekly

Track competitor changes with strategic implications.

Roadmap Alignment

On-demand

Generate alignment memos with options and trade-offs.

PRD Draft

On-demand

Draft PRDs grounded in customer evidence and context.

See It In Action

Try all six jobs in the interactive demo with a complete mock enterprise dataset.