Private AI infrastructure · Brisbane
Context should
survive the chat.
Cloud Memory gives different AI tools one durable place to recall decisions, understand active projects and leave useful work behind for the next conversation.
- 01
- Private by default
- 02
- Portable across models
- 03
- Evidence stays attached
01 / The problem
A good conversation should not become lost infrastructure.
AI tools are brilliant within a chat and forgetful between them. Decisions drift, tasks lose their source, and every new model starts by asking what already happened. Cloud Memory turns that scattered history into a small, governed context layer.
One private edge, many clients
The system, end to end.
Canonical records stay in one owner-scoped store. Search indexes and Obsidian notes are rebuildable projections, not competing sources of truth.
Memory with a lifecycle
Durable memories and standing directives carry importance, source, status and retrieval evidence. Superseded context stays auditable instead of silently disappearing.
Projects with receipts
Tasks show status, priority, model, client and the original chat. Work can be completed, archived or reopened without losing the record of how it got there.
A longer view
Roadmaps hold AI-suggested next steps beyond launch. Ideas remain separate from committed tasks until a human deliberately promotes them.
Bounded recall
Hybrid search combines exact, semantic, temporal and entity signals. A small context brief gives an AI what matters without loading the entire archive.
Useful archives
Archived work remains browsable for the owner while routine AI context ignores it. That keeps old incomplete tasks from masquerading as current priorities.
Human-readable projection
n8n projects curated summaries into Obsidian through pCloud. Those notes travel to every device while the canonical database remains protected.
The intelligent hand-off
Recall. Work. Capture.
The workflow is deliberately bounded. Models recall when prior context matters, update explicit tasks while working, and capture only the few durable facts worth carrying forward.
- 01Context brief
Start with the smallest useful brief
Active directives, relevant memories, current projects and outstanding tasks arrive together, with archive noise excluded.
- 02Task start
Keep provenance attached to the work
The active task records the real client, model and chat URL, so progress can be traced back to the conversation that created it.
- 03Finish + capture
Leave a precise, durable hand-off
Known outcomes close the task. Only concise decisions, corrections or standing preferences are proposed as memory.
Designed for the owner
A command centre, not a data dump.
The private dashboard makes project health, outstanding tasks and memory provenance legible on desktop and mobile without exposing any of that data here.
PROJECT STATUSWork, with receipts.
ACTIVEProject Atlas4 outstanding
Outstanding4In progress2Blocked1Complete7
IN PROGRESSVerify the mobile workflow
PLANNEDPublish the reference guide
Illustrative interface with fictional project data. No private Cloud Memory records are published on this site.
The boundary matters
The product is private.
The architecture can be public.
This page explains what Cloud Memory is without exposing its memories, directives, tasks, project names, operational identifiers or source conversations.
- Authentication
- GitHub owner identity and scoped MCP OAuth
- Canonical data
- Owner-scoped rows in Cloudflare D1
- Derived systems
- Rebuildable vectors, notes and encrypted exports
- Public surface
- Architecture and fictional examples only
Cloud Memory · Drew's Digest
One memory layer.
Every useful model.
Explore Drew's Digest ↗
The workspace itself is owner-only. This public reference remains deliberately read-only.