Plan · Govern · Hand off — the planning factory in front of the AI execution agents

Plans, governed before a single line is built.

Hand PFactory a project plan — uploaded as docx / pdf / markdown, or via the MCP control plane, the CLI, or a GitHub issue. PFactory enriches it with your live org and cloud context, decomposes it, runs architecture / security / best-practice / feasibility gates, and emits governed GitHub epics + child issues that AIFactory executes.

How it works →   Why PFactory →   Roadmap →

The planning pipeline

plan → governed epics
01

Ingest

docx · pdf · md
02

Enrich

Backstage · K8s · cloud
03

Decompose

epic + children
04

Review & approve

gates + human
05

Emit

GitHub → AIFactory

Part of the Factory family — a governed, verified, observable autonomous software factory. PFactory plans · AIFactory builds · TFactory verifies · CFactory watches over all four. → Why Factory

See it in action

Ingest a plan → process → review across architecture · security · best-practice · feasibility lenses → red-team / adversarial pass → board → approve → emit tagged GitHub epics + issues. Captured from the live portal.

Plans on the board Plans in the list view A plan flagged needs-attention A verified Code-aware project index MCP servers per agent

See the full portal tour →

Why a planning factory

The 2026 coding agents — Devin, Factory’s droids, the Copilot coding agent — collapsed planning into execution: hand them a vague ask and they plan and build in one loop. That optimizes generation speed. But the real constraint is the opposite: the gap between what agents generate and what teams can confidently approve. PFactory makes the plan the reviewed, auditable artifact — so you get AI velocity with human-grade governance, grounded in your real infrastructure.

What PFactory does

Plug in anything

PFactory is extensible by design. A declarative registry lets you add MCP servers, skills, agents, and templates without forking. Cloud and IaC best-practice MCP servers (AWS · Azure · GCP · Terraform) drive automatic review. Templates are Backstage Software Template-compatible and carry their own embedded policy rules — a gcp-project template scaffolds the project and enforces its org policies, regions, and IAM baselines. As the clouds and best practices change, PFactory watches them and proposes template updates via pull request — never silent edits.

It also speaks GitHub’s native agentic surface: free GitHub Models inference (github-models/openai/gpt-4.1), copilot:delegate to hand a draft to the Copilot cloud agent, and an HTTP MCP server (POST /mcp) that serves the epic, requirements, decomposition, signed Task Contract v2, and review status to whatever agent builds from the plan. All opt-in, all additive.

How it fits

PFactory is the third factory in the suite — plan → build → test, governed end to end:

One task, three factories

plan → build → test

PFactory

Plans & governs

plan

AIFactory

Builds the code

code

TFactory

Verifies with tests

test
Factory Role
PFactory Plans & governs — ingest, enrich, decompose, review, emit issues
AIFactory Executes — spec → plan → code → QA on the emitted issues
TFactory Tests — generates & runs tests on the built code

Documentation

Tracking