Brain-like organic memory for AI agents and IDE assistants — powered by QubicDB.
Includes ready-to-use configurations for Windsurf, Cursor, VS Code (GitHub Copilot), and Claude Code. Also available as a Claude Code Plugin via marketplace.
Option A: Non-vector Docker Compose
docker compose -f docker-compose.qubicdb.yml up -dOption B: Vector Docker Compose
Clone qubicdb and skills side-by-side:
your-folder/
├── qubicdb/
└── skills/
Then run:
docker compose -f docker-compose.qubicdb.vector.yml up -d --builddocker-compose.qubicdb.vector.yml expects the vector model at ../qubicdb/dist/MiniLM-L6-v2.Q8_0.gguf.
Important: Download the GGUF model first. Full download commands are documented in SETUP.md.
Helper instructions and local patch scripts are also available in the main qubicdb repo:
- patches/windows-local-vector
- patches/linux-local-vector
- patches/macos-local-vector
- patches/vector-wrapper
Option C: Bundled Vector Docker Compose
docker compose -f docker-compose.qubicdb.bundled.yml up -dThis lane uses qubicdb/qubicdb-bundled:latest, which includes the example MIT-licensed MiniLM-L6-v2.Q8_0.gguf model (about 25 MB). No local build or dist mount is required.
Option D: Docker Run (non-vector)
docker pull qubicdb/qubicdb:latest
docker pull qubicdb/qubicdb-ui:latest
docker network create qubicdb-net
docker run -d \
--name qubicdb \
--network qubicdb-net \
-p 6060:6060 \
-v qubicdb_data:/app/data \
-e QUBICDB_HTTP_ADDR=:6060 \
-e QUBICDB_DATA_PATH=/app/data \
-e QUBICDB_ADMIN_ENABLED=true \
-e QUBICDB_ADMIN_USER=admin \
-e QUBICDB_ADMIN_PASSWORD=changeme \
-e QUBICDB_MCP_ENABLED=true \
-e QUBICDB_MCP_PATH=/mcp \
-e QUBICDB_MCP_STATELESS=true \
-e QUBICDB_MCP_RATE_LIMIT_RPS=30 \
-e QUBICDB_MCP_RATE_LIMIT_BURST=60 \
-e QUBICDB_MCP_ENABLE_PROMPTS=true \
-e QUBICDB_MCP_API_KEY=qubicdb-mcp-secret-key \
-e QUBICDB_ALLOWED_ORIGINS=http://localhost:6060,http://localhost:8080,http://127.0.0.1:8080 \
-e QUBICDB_REGISTRY_ENABLED=false \
qubicdb/qubicdb:latest
docker run -d \
--name qubicdb-ui \
--network qubicdb-net \
-p 8080:80 \
qubicdb/qubicdb-ui:latest📖 For full non-vector setup, vector setup, vector parameters, and troubleshooting, see SETUP.md.
Verify:
curl http://localhost:6060/healthAdmin UI: http://localhost:8080 — login with admin / changeme.
Note: The
QUBICDB_MCP_API_KEYmust match theX-API-Keyheader in your IDE's MCP config. Change both if you use a custom key.
Option A: Claude Code Plugin (recommended for Claude Code)
/plugin marketplace add qubicDB/skills
/plugin install qubicdb-skills@qubicdb-agent-skillsOption B: Copy IDE folder — pick your IDE below, copy the relevant folder into your project root.
Start a conversation — the AI will use QubicDB as persistent memory automatically.
Copy .windsurf/ into your project root. Add MCP server in Windsurf settings (or use mcp_config.json as reference):
your-project/
└── .windsurf/
├── rules/
│ └── qubic.md ← always-on rules (191 lines)
├── skills/
│ └── qubic/
│ └── SKILL.md ← skill definition (140 lines)
└── workflows/
├── qubic-init.md ← /qubic-init workflow
├── qubic-search.md ← /qubic-search workflow
└── qubic-write.md ← /qubic-write workflow
MCP config for Windsurf settings:
{
"mcpServers": {
"qubicdb": {
"serverUrl": "http://localhost:6060/mcp",
"headers": {
"X-API-Key": "qubicdb-mcp-secret-key"
}
}
}
}Copy .cursor/ into your project root:
your-project/
└── .cursor/
├── mcp.json ← MCP server config
└── rules/
├── qubic-rules.mdc ← always-on rules (full port of Windsurf rules)
├── qubic-skill.mdc ← skill definition (full port of Windsurf skill)
├── qubic-init.mdc ← init workflow
├── qubic-search.mdc ← search workflow
└── qubic-write.mdc ← write workflow
Copy .vscode/ and .github/ into your project root:
your-project/
├── .vscode/
│ └── mcp.json ← MCP server config
└── .github/
└── copilot-instructions.md ← all rules + skills + workflows combined
Copy .claude/ into your project root:
your-project/
└── .claude/
├── CLAUDE.md ← project rules (always loaded)
├── settings.local.json ← MCP server config
└── skills/
├── qubic/
│ └── SKILL.md ← main skill definition
├── qubic-init/
│ └── SKILL.md ← /qubic-init skill
├── qubic-search/
│ └── SKILL.md ← /qubic-search skill
└── qubic-write/
└── SKILL.md ← /qubic-write skill
| IDE | MCP Config | Rules/Skills |
|---|---|---|
| Windsurf | mcp_config.json (add in settings) |
.windsurf/rules/ + .windsurf/skills/ + .windsurf/workflows/ |
| Cursor | .cursor/mcp.json |
.cursor/rules/*.mdc (5 files) |
| VS Code | .vscode/mcp.json |
.github/copilot-instructions.md |
| Claude Code | .claude/settings.local.json |
.claude/CLAUDE.md + .claude/skills/ (4 skills) |
| Content | Description |
|---|---|
| Rules | Always-on behavior: session lifecycle, forced behaviors (no fabrication, auto-write triggers), persona, naming conventions |
| Skill | Architecture (index/thread/neuron model), MCP tool reference, metadata types, usage examples |
| Init workflow | Register project brain index, load existing memories |
| Search workflow | Spreading activation search, metadata filters, parameters |
| Write workflow | Store decisions, preferences, todos, facts with metadata |
QubicDB gives your AI assistant a persistent, organic memory across conversations:
- One brain per project —
brain-{project}index, created once, reused forever - One thread per conversation —
conv-{uuid}groups neurons within a session - Hebbian learning — memories that fire together, wire together
- Spreading activation search — finds related memories through synapse connections
- Lifecycle states — Active → Idle → Sleeping → Dormant with automatic consolidation
This repo is a Claude Code Plugin Marketplace. Install directly:
# Add the marketplace
/plugin marketplace add qubicDB/skills
# Install core skills (MCP memory: write, search, recall, context)
/plugin install qubicdb-skills@qubicdb-agent-skills
# Install experimental skills (multi-index, agent-to-agent, sentiment, REST API, RAG)
/plugin install qubicdb-experimental@qubicdb-agent-skillsPlugin structure:
plugins/
├── qubicdb-skills/ ← core plugin
│ ├── .claude-plugin/plugin.json
│ ├── .mcp.json ← auto-configures MCP server
│ └── skills/
│ ├── qubic/SKILL.md
│ ├── qubic-init/SKILL.md
│ ├── qubic-search/SKILL.md
│ └── qubic-write/SKILL.md
└── qubicdb-experimental/ ← experimental plugin
├── .claude-plugin/plugin.json
└── skills/
├── multi-index-research/SKILL.md
├── agent-to-agent/SKILL.md
├── conversation-chains/SKILL.md
├── sentiment-journal/SKILL.md
├── knowledge-base-server/SKILL.md
└── rag-context-assembly/SKILL.md
QubicDB now supports global and multi-index operations via MCP:
| Tool | Description |
|---|---|
qubicdb_list_indexes |
List all registered indexes with stats |
qubicdb_global_search |
Search across ALL active indexes |
qubicdb_multi_search |
Search across a specific list of indexes |
qubicdb_recent_indexes |
Get most recently active indexes |
Use case: Agents can now discover relationships across hundreds of repositories/indexes — perfect for monorepo insights and cross-project memory.
The qubicdb-experimental plugin showcases diverse use cases beyond IDE memory:
| Skill | Description |
|---|---|
| multi-index-research | Separate brains per research domain, cross-reference between them |
| agent-to-agent | Multiple agents sharing memory through a common brain with role metadata |
| conversation-chains | Track multi-session conversations with thread_id chaining and parent links |
| sentiment-journal | Emotion-aware journaling leveraging built-in VADER sentiment analysis |
| knowledge-base | Build a persistent knowledge base with rich metadata and hybrid search |
| rag-context-assembly | Token-budgeted context assembly using spreading activation (not just vector similarity) |
Each skill includes full Prerequisites with Docker setup and MCP config for all 4 IDEs.
- QubicDB Server: github.com/qubicDB/qubicdb
- Admin UI: github.com/qubicDB/qubicdb-ui
- Docker Hub: hub.docker.com/r/qubicdb/qubicdb
- Website: qubicdb.github.io/qubicdb-web
- API Docs: qubicdb.github.io/docs