GitHub Strategy

Status: WORKING INTERNAL STRATEGY
Date: 2026-05-09
Applies to: public github.com/pgmnemo/pgmnemo surface
Not for publication: internal operating document


1. Objective

Use GitHub as the primary trust surface for pgmnemo, so the repository itself communicates:

  1. technical maturity
  2. operational discipline
  3. credible differentiation
  4. maintainer responsiveness
  5. installability and proof

The target is:

a skeptical PostgreSQL engineer can open the repo, understand what pgmnemo is, install it, see evidence that it works, and believe that this project might become the default agent-memory extension for PostgreSQL.


2. Strategic position

pgmnemo should not look like a generic AI tool. It should look like a serious PostgreSQL extension with a differentiated AI-memory thesis.

Target feel: - pgvector on installation clarity - Apache AGE on identity and docs seriousness - pgvectorscale on benchmark-backed claims - TimescaleDB on onboarding polish and confidence

Avoid looking like: - a research dump - a founder notebook - a loose internal dogfood repo - a roadmap-heavy project with weak operator proof


3. Best owner model

Best single owner

If one agent must own the GitHub surface, the best fit is:

growth_lead

Reason: - explicitly owns README, docs, issue templates, positioning, launch, community, and competitive narrative

Required co-owners

GitHub must not be run by growth_lead alone.

Working model: - Growth Lead — public surface owner - Technical Lead — release hygiene owner - Chief Architect — technical claim / compatibility truth owner - Startup Mentor — external critique owner - Competitive Intelligence / Literature Scout — competitor watch inputs

This split avoids both failure modes: 1. technically correct but externally unreadable 2. externally polished but internally dishonest


4. What GitHub must prove

Every GitHub-facing change should strengthen one of these proof classes:

P1. Install proof

GitHub must answer: - supported PostgreSQL versions - supported pgvector versions - Docker path - native path - upgrade path - what is experimental vs stable

P2. Product proof

GitHub must answer: - what pgmnemo is - why it is different from mem0, Zep, Constructive, and generic vector stacks - who it is for - what exact SQL/API surface is stable today

P3. Performance / quality proof

GitHub must answer: - benchmark methodology - benchmark datasets - recall / latency / footprint claims - what has been measured vs inferred

P4. Maintenance proof

GitHub must answer: - release cadence - issue templates - bug triage expectations - contribution flow - upgrade discipline

P5. Strategic proof

GitHub must answer: - why this belongs in PostgreSQL - why provenance-gated memory matters - why this is a long-term extension product rather than an internal experiment


5. External benchmark lessons

pgvector

Copy: - ruthless installation clarity - many environment paths - examples before philosophy

Apache AGE

Copy: - strong project identity - official documentation anchor - visible seriousness and ecosystem fit

pgvectorscale

Copy: - benchmark-backed claims - clear differentiation - contributor path separated from user path

TimescaleDB

Copy: - explicit quickstart - obvious next steps - confidence-building onboarding

Constructive

Copy: - ecosystem framing and getting-started momentum

Do not copy: - broad platform framing that weakens extension-trust focus


6. Strategic rules

Rule 1 — GitHub is a product surface

Treat README, releases, docs, examples, issue templates, and benchmarks as product components.

Rule 2 — Public contract beats internal progress

If main is ahead of the latest release, public docs must still reflect one coherent reality. No split-brain between: - README - release notes - CHANGELOG.md - META.json - extension control file - docs examples

Also decide one explicit release model and document it: - releases are canonical - or tags are canonical

Rule 3 — Installation clarity beats feature count

For early adoption, one clean install path is more valuable than many extra features with unclear setup.

Rule 4 — Benchmarks must be reproducible or omitted

Any performance or recall claim on GitHub must link to: - dataset - query protocol - environment - command path - exact metric definition

Rule 5 — Stable path first, roadmap second

Public docs must lead with: - what works today - how to install it - how to use it safely

Rule 6 — Every public page must know its audience

  • README = first-time evaluator
  • INSTALL = operator
  • USAGE = implementer
  • MIGRATION = adopter
  • benchmarks = skeptic
  • CONTRIBUTING = contributor

Rule 7 — Visible maintenance is part of the product

Issue templates, response rhythm, release discipline, compatibility notes, CODEOWNERS, SECURITY.md, and CODE_OF_CONDUCT.md are part of the adoption contract.

Rule 8 — Technical honesty compounds

Document known limitations directly: - supported PG versions - unsupported environments - performance caveats - RLS caveats - non-goals

Rule 9 — Public repo language must match the extension form factor

Avoid stale language from the pre-extension phase: - install.sql - mem.* wedge-first framing - internal dogfood jargon - paper-program language as primary product explanation

Rule 10 — The repo should look smaller and sharper, not bigger and fuzzier

pgmnemo should look like a focused extension with a hard technical thesis.

Rule 11 — README above-the-fold must route the first four questions

The first visible screen should route a newcomer to: - Quickstart - Install / Upgrade - Examples - Roadmap / Status


7. Success condition

GitHub is strategically healthy when:

  1. a PostgreSQL engineer can install pgmnemo in under 10 minutes
  2. a reviewer can tell exactly which version is latest and what changed
  3. a skeptical adopter can distinguish stable surface from planned surface
  4. a contributor can see how to file issues, propose changes, and run tests
  5. a competitor reviewer must concede that the repo looks technically serious

8. Decision

pgmnemo GitHub will be run as a trust-building technical product surface.

Permanent owner: growth_lead
Permanent co-owner: technical_lead
Required gate reviewers: chief_architect, startup_mentor