Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@DGUI-HyperMem Remember that I prefer dark mode in all my tools."
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
DGUI-HyperMem (DeckerGUI HyperMemory) — a self-hosted, hybrid long-term memory MCP server for AI agents, running entirely on Cloudflare Workers. Retrieval fuses vector search with full-text search, then a JEV reasoning layer (Choice / Noul / Score) re-ranks and curates what gets remembered.
A DeckerGUI project.
Stores durable memories — an LLM/JEV layer assigns each memory a type, a salience score and a durability judgement; non-durable chatter is dropped.
Recalls hybrid — Vectorize ANN + D1 FTS5 BM25 candidates, fused by reciprocal rank, then re-ranked by JEV so the best memory wins even when wording differs.
Self-maintains — near-duplicate/contradicting memories are superseded automatically.
Gets better with use — every JEV decision is logged and flushed to a HuggingFace dataset, building a training corpus for the reasoning layer.
┌───────────────────────────── Cloudflare Worker ─────────────────────────────┐
MCP client │ │
(opencode, etc.)│ /mcp ─ Streamable HTTP MCP add · search · list · profile · forget │
───────────┼─► /api/* ─ REST mirror · help · sync_jev_dataset · jev_queue_stats│
│ │ │
│ ┌─────────────────┴─────────────────┐ │
│ │ JEV layer │ │
│ │ Choice · Noul · Score (systemone)│ │
│ └─────────────────┬─────────────────┘ │
│ ┌───────────────┬───────────────┴───────────┐ │
│ D1 (SQLite) Vectorize (768d) Workers AI │
│ memories + FTS5 embeddings fallback model │
│ └───────────────┴───────────────┬───────────┘ │
│ jev_examples queue │
│ │ hourly cron │
└────────────────────────────────────────────┼───────────────────────────────┘
▼
🤗 ctaxnagomi/DGUI_HYPERMEM-JEV (training brain)Related MCP server: MCP Memory
Layer | Choice |
Runtime | Cloudflare Workers (ES modules, |
Protocol | MCP Streamable HTTP (stateless, JSON responses) via |
Database | Cloudflare D1 — |
Vectors | Cloudflare Vectorize — |
Reasoning | TypeSafe AI Jev / System One ( |
Route | Method | Purpose |
| POST | MCP (Streamable HTTP). Requires auth. |
| POST | REST mirror of the tools |
| POST | Flush queued JEV examples to the dataset now |
| GET | Queue status (pending / uploaded / error) |
| GET | Unauthenticated status |
MCP tools: add, search, list, profile, forget, help, sync_jev_dataset,
jev_queue_stats.
Auth accepts Authorization: Bearer <MCP_TOKEN>, x-api-key: <MCP_TOKEN>, or ?token=.
npm install
# 1. Create resources (once)
npx wrangler d1 create dgui-hypermem
npx wrangler vectorize create dgui-hypermem --dimensions=768 --metric=cosine
# 2. Point wrangler.jsonc at your D1 id, then apply migrations
npx wrangler d1 migrations apply dgui-hypermem --remote
# 3. Secrets
npx wrangler secret put MCP_TOKEN # bearer token clients must present
npx wrangler secret put TYPESAFE_API_KEY # optional: enables the TypeSafe JEV backend
npx wrangler secret put HF_TOKEN # optional: enables dataset sync
# 4. Ship
npx wrangler deploywrangler.jsonc vars)Var | Default | Meaning |
|
|
|
|
| TypeSafe model alias |
|
| Workers AI fallback |
|
| Embedding model |
|
| Default memory namespace |
|
| Training-brain dataset |
{
"mcp": {
"dgui-hypermem": {
"type": "remote",
"url": "https://dgui-hypermem.<your-subdomain>.workers.dev/mcp",
"enabled": true,
"headers": { "Authorization": "Bearer {env:DGUI_HYPERMEM_TOKEN}" }
}
}
}Every JEV decision is recorded as an instruction row — the exact state and typed
questions that were sent, and the answers that came back — in the jev_examples table.
An hourly cron (17 * * * *) and the sync_jev_dataset tool append pending rows to
ctaxnagomi/DGUI_HYPERMEM-JEV,
updating train.jsonl and running totals in metadata.json.
|
| Recorded when |
|
| A memory is stored/typed |
|
| Recall re-ranks candidates |
|
| A contradiction check runs |
src/
index.ts MCP server, REST routes, scheduled handler
store.ts add / search / list / profile / forget (hybrid retrieval)
jev.ts JEV layer: analyze, rerank, supersede (TypeSafe + Workers AI)
dataset.ts jev_examples queue + HuggingFace flush
types.ts shared types
util.ts ids, hashing, base64, timing-safe compare
migrations/ 0001_init.sql, 0002_jev_examples.sql
wrangler.jsonc bindings, vars, cronDGUI-HyperMem is a DeckerGUI project.
Who | Contribution | Link |
TypeSafe AI | Jev — the first System One model — and the Choice / Noul / Score primitives the reasoning layer is built on. | |
DeckerGUI | Design, implementation and operation. | |
KrackedDevs | Community credit and support. | KrackedDevs |
CTECX | Knowledge / corpus partner. | CTECX |
Jev, System One, and the Choice / Noul / Score primitives are TypeSafe AI's and are used under their MIT-licensed public documentation (https://docs.typesafe.ai).
Contributions are welcome. See CONTRIBUTING.md; contributors are listed in CONTRIBUTORS.md.
MIT © 2026 DeckerGUI.
This server cannot be deployed
Persistent memory for AI agents. Search and store durable facts, preferences and decisions.
Persistent cloud memory for AI agents. Store and search key-value memories across sessions.
Persistent memory for AI agents. Semantic search, memory graph, W3C DID identity.
Hosted persistent memory with semantic search, importance and TTL for AI agents.
