01 / MEMORY
Find the right context, with reasons.
Temporal queries, supersession, labels, archive, bounded boosts and explainable ranking keep durable knowledge useful instead of merely large.
OPEN SOURCE · SELF-HOSTED AT THE EDGE
One memory.
Every model.
Your cloud.
Give every approved AI client the same durable context, project state and next action. No home server. No central Cloud Memory account. No n8n required.
- Cloudflare Worker
- D1 canonical
- GitHub OAuth
- Remote MCP
01Codex
02Claude
03OpenCode
04Web
CANONICALCMD1 / OWNER LOCKED
AI conversations are temporary.
Your decisions, directives and unfinished work should not be.
THE WORKSPACE
Memory that knows
work has a lifecycle.
Retrieval is only half the problem. Cloud Memory connects what your models know with what they are doing next.
02 / PROJECTS
A board models can finish.
Tasks carry client, model, source chat, attention signals and completion evidence. Done work ages into history without crowding the board.
03 / ROADMAPS
Keep the next horizon visible.
Separate future enhancements from immediate work. Models can suggest ideas, while promotion into the task board stays explicit and traceable.
04 / PORTABILITY
Canonical, not captive.
Encrypted exports and managed Obsidian Markdown create useful recovery and reading layers without turning them into competing writers.
- D1Canonical truth
- GitHubEncrypted recovery
- ObsidianReadable projection
THE BOUNDARY
Public code.
Private workspace.
Every adopter gets an isolated stack. This project has no central data plane and no path into your deployment.
YOUR AI CLIENTSCodex · Claude · OpenCode · compatible web clients
OAuth-protected MCP
YOUR CLOUDFLARE ACCOUNTWorker · dashboard · D1 · KV
Optional, derived and replaceable
SEMANTICVectorize + Workers AI
PORTABILITYYour GitHub + WebDAV
Immutable numeric GitHub ID, not a public sign-up surface.
Useful without AI calls; semantic retrieval turns on only when you choose.
Connected, authenticated and verified are recorded as separate states.
START WITH TWO ACCOUNTS
Your memory stack in five deliberate steps.
The deployment guide walks from an empty account to authenticated MCP canaries without collecting your secrets.
Open the complete guide ↗- 01Deploy
Create the Worker and resources in your Cloudflare account.
- 02Lock
Allowlist your immutable numeric GitHub user ID.
- 03Authorise
Register one exact GitHub OAuth callback.
- 04Connect
Add the same remote MCP endpoint to each client.
- 05Verify
Canary health, board, search and one approved write.
CLEAR ANSWERS
Before you deploy.
Is it free?
The software is MIT licensed. The default avoids Workers AI and Vectorize calls, but Cloudflare and GitHub allowances can change. Review current provider pricing before deployment; the project never enables billing for you.
Does it need n8n or a home server?
No. The core needs only your Cloudflare and GitHub accounts. n8n, WebDAV, Obsidian, semantic search and repository backups are optional integrations.
Can the maintainers see my memories?
No. There is no shared Cloud Memory service. Your deployment and data live in your provider accounts. You remain responsible for provider settings and security.
Why is semantic search off by default?
Lexical and temporal retrieval is useful, predictable and cost-conservative. Enable semantic search when you want it, then backfill pending vectors from canonical D1 records.
COMMUNITY EDITION · V1.0