Case study / 02
AI Coding-Agent
Memory System
A governed memory operating system for AI coding agents: durable project knowledge, task state, retrieval, and evidence-based handoffs.
Type
AI memory system
Canonical truth
Git-reviewed Markdown
Status
Built internal system
Stack
GBrain · Postgres/pgvector · Beads
Overview
AI coding agents lose context, repeat mistakes, and mix temporary task state with durable knowledge.
memory-os solves that with a governed system where agents keep durable project knowledge separate from in-flight task state, backed by retrieval and evidence rules.
System architecture
Knowledge
Git-reviewed Markdown
Retrieval
GBrain
Projection
Postgres/pgvector
Task state
Beads
Skills
Agent skills
Handoff
Verification-gated
Problem
Coding agents drift: context is lost between sessions, mistakes repeat, and temporary state pollutes durable knowledge.
What it does
- Durable project knowledge
- Evidence / date / source / owner rules
- Setup / verify / sync workflows
- Backup / restore workflows
- Verification-gated handoffs
What this proves
- AI-assisted engineering operations at scale.
- Memory governance for coding agents.
- Postgres/pgvector usage for retrieval.
- Agent workflow architecture and evidence-based automation discipline.
Stack
- Git-reviewed Markdown as canonical truth
- GBrain retrieval
- Postgres / pgvector projections
- Beads task state and agent skills