Vaultic MCP
Linux上のMCP互換AIアシスタント向けローカルメモリ。プロジェクト情報を保存・取得し、エージェントが開けるデスクトップビューアーで記録を確認し、修正履歴を保持できます。埋め込みモデルを同梱。依存関係のインストール後はオフラインで動作します。料金は一度限りの€135で、Vaulticの週額・月額・年額サブスクリプションはありません。Morrisonsのクラウドアカウントやコールバックも不要です。
ドキュメント
What does that mean in practice?
Suppose you and your AI assistant plan a website today. You save the pages you agreed on and the next job. In a later chat, you can ask, “Where did we leave off with the website?” The assistant can look up the saved information and help you carry on.
You choose what to save. Automatic conversation saving needs a working connection to your AI app; the included instructions help you set this up.
Pick up an earlier project
Ask about earlier work, decisions and preferences in a new chat. Look up where you left off and what comes next.
Open your memory
Use the desktop app to read saved notes and conversations, view your assistant’s settings and see links between people, projects and ideas. Your AI assistant can open the app for you.
Get help with setup
Give the included Agent Manual to your AI assistant. It can help connect Vaultic and set up how information is saved and looked up. Your AI app may ask you to approve some steps.
“What were we working on earlier?”
Once Vaultic is connected, ask in your own words. Your AI assistant uses the saved information to help answer.
- 1
Save the useful context
Work on a project and save its details, decisions and next step. You can also set up automatic saving of conversations. - 2
Start a fresh session
Open a new chat with the same memory connection and assistant settings. Ask about the earlier project. - 3
Check it and continue
Read the earlier details and continue working. Open the app to check saved records and correct anything that needs updating.
Example workflow. Recall depends on what was saved, the query and your client setup; it is selective and can miss information.
Your memory. On your machine.
Local storage
Memories and supporting records stay in local databases and stores. No required Vaultic cloud account, telemetry or licence server.
Encrypted content
Supported content fields are encrypted at rest with AES-256-GCM by default. Keys are kept separately. Metadata and search-index privacy limits are explained below.
Search in your own words
Semantic retrieval works alongside encrypted keyword search. Word-form matching and prefixes help find relevant saved text.
More than a chat log
Keep distilled memories, source transcripts, a knowledge graph, profiles and an evidence-linked relationship ledger.
Corrections with history
Supersede an outdated memory or refute a ledger claim while keeping its history. Forgetting archives a memory rather than silently erasing it.
Choose your workflow
Use MCP over stdio or configured loopback HTTP. Included reference hooks cover Claude Code; other hosts need their own supported configuration.
Download. Connect. Make it yours.
- Download Vaultic 0.1.5 from your Gumroad purchase. The package includes the embedding model, Agent Manual and reference hooks.
- Install using the included instructions. Use Python 3.12+ with SQLite 3.51.3+. The supplied Linux instructions use
uvand Python 3.13; check the actual SQLite version before starting. - Give the manual to your agent. Let it help connect Vaultic and configure capture and recall for your host. Apply any required host approvals.
- Try a fresh-session recall. Save a harmless test fact, open a new session and ask for it. Verify the result before relying on the workflow. MCP configuration example
Use the real absolute path to the executable inside your installed environment. Register it at user/global scope if you want it available across projects. Client configuration formats vary; this example is for clients using mcpServers.
{
"mcpServers": {
"vaultic": {
"command": "/home/YOUR-NAME/vaultic-env/bin/vaultic-mcp"
}
}
}
Registration makes the tools available. Automatic capture and recall also require a working host integration.
Once installed, the bundled default embedding model supports local offline operation. An external embedding API or a different downloaded model is optional. Your AI client's own connection requirements still apply.
A complete memory toolkit, a compact daily interface.
Vaultic 0.1.5 exposes 111 MCP tools in its full profile. The default profile advertises 35, including two tools for discovering and calling the rest. Your client controls how tool descriptions enter its context.
remember and recall
Save useful facts and find them again through semantic and encrypted keyword search.
retrieve_context
Bring relevant memories, graph context and available source evidence into the current task.
open_gui
Let your agent open the desktop viewer so you can inspect your local memory.