Your AI forgets everything.
Fix that.

SubVault is an MCP server that gives Claude, Cursor, and Copilot permanent memory. Two words — remember and vault — and your AI never starts from zero.

No credit card. No limits. Free during early access.

Built for conversations

Every AI conversation contains decisions, ideas, and context that vanishes when the session ends. SubVault captures what matters and brings it back when you need it.

Vault a research session

Say "vault this" and key decisions, facts, and action items are extracted and saved.

Remember across sessions

Say "remember what I decided about pricing" and your AI has full context instantly.

Built for developers

SubVault remembers code decisions, not just conversations. Architecture patterns, API choices, debugging solutions — searchable across every session, every tool.

Vault a project session

Every architectural choice is saved with reasoning, searchable across tools.

Remember project context

Your AI knows the codebase decisions before you write a line.

Set up in 30 seconds

Create an account at subvault.ai. Takes 10 seconds.

Connect your AI tools

Run the installer, double-click the setup file, or paste config manually.

Say remember or vault

Your AI memory builds with every conversation.

Terminal

Download

Manual

Run one command. Configures Claude Desktop, Cursor, and VS Code automatically.

$ curl -fsSL https://subvault.ai/setup.sh | bash -s YOUR_API_KEY

Download the setup file. Double-click it in Finder. Everything is configured automatically.

setup_subvault.command

Your API key is already included. Just double-click.

Copy the config for your AI tool. Your API key is pre-filled.

Claude Desktop

~/Library/Application Support/Claude/claude_desktop_config.json

{ }

Cursor

~/.cursor/mcp.json

VS Code + Copilot

.vscode/mcp.json — uses "servers" key, not "mcpServers"

Not a chat log. An intelligence layer.

Most tools dump raw transcripts. SubVault extracts structured knowledge — scored by authority, recency, and relevance — so the right context surfaces every time.

Decisions + reasoning

"Chose PostgreSQL over MongoDB — billing needs relational integrity"

People + relationships

"Sarah owns frontend, Mike handles DevOps, report to Lisa"

Action items + status

"Update billing webhook before Friday — blocked on Stripe keys"

Project context

"API at v2, migration from v1 in progress, 3 endpoints left"

Raw chat history

Dumps entire transcripts

No relevance ranking

Fills the context window

Same info every time

Locked to one AI tool

SubVault extraction

Extracts facts, decisions, patterns

Scores by authority + recency + relevance

Token budgeted assembly

Adapts context to your query

Works across every MCP client

What happens under the hood

Structured extraction

Conversations become self contained facts, decisions, code patterns, people, and action items. Each stands alone 6 months later.

4 axis scoring

Ranked by authority (you said it > AI extracted), recency, relevance to your query, and salience.

Query classification

"Who is Sarah?" surfaces people. "What's the status?" surfaces actions. Context adapts automatically.

Token budgeted assembly

Fixed token budget across sections. No context window bloat. Only the most relevant items make it in.

Dedup + contradiction detection

Content-hash dedup prevents duplicates. Contradictions flagged when new decisions override old ones.

Injection protection

Every vault item is sanitized before entering AI context. Prompt injection patterns detected and stripped.

Pricing

Early access

Free during early access

Full access to everything. Early access users are founding members, grandfathered when pricing starts.

Your AI forgets everything.Fix that.