A password manager for your AI memory

One memory.
Every AI.

Memobase is a private, portable memory vault for your AI. Tell Claude something today and ChatGPT knows it tomorrow — your context follows you across every assistant, owned by you, not locked inside one vendor.

Connect any AI in seconds One URL

Add Memobase as a connector in Claude or ChatGPT, sign in once, and your vault is shared across both. Import your existing chat history to give every AI an instant head start.

Create free vault See how it works

Start free View plans

The Challenge

Break the Amnesia Loop.

Context windows are expanding, but they remain ephemeral. Every new session is a reset—your agent forgets your architecture, your style, and your progress.

Context Erosion

As conversations grow, older facts are silently evicted from the context window to make room for new ones.

Session Isolation

Great ideas from yesterday's chat are inaccessible today. You're stuck in a loop of re-explaining your setup.

Fragmented Identity

Claude knows one thing, ChatGPT knows another. There is no shared "source of truth" for your AI toolchain.

Static Intelligence

Without long-term memory, agents can't evolve with you. They remain generic assistants rather than specialized partners.

Core Architecture

Built for production

🌙

The "Dream Phase"

Background distillation that turns noisy session logs into durable, high-signal rules and project insights.

🪝

Passive Intelligence

Deterministic HTTP hooks capture context automatically. The agent learns without "remembering to remember."

🔍

Hybrid Graph Retrieval

Combines vector similarity with Knowledge Graph relationships for precise, context-aware recall.

🔌

MCP-native protocol

Built on the Model Context Protocol. Works natively with Claude, ChatGPT, and Cursor.

🔐

Privacy by Design

PostgreSQL Row-Level Security ensures your memory is your own. Self-custody options on the roadmap.

📊

Insight Dashboard

Visualize your knowledge graph and vector space. Manage API keys and monitor usage in real-time.

Implementation

Up and running in minutes

⚙️

Initialize Core

Add the Memobase MCP server to your AI configuration with a single JSON entry.

🔒

Neural Handshake

Authenticate once via secure OAuth. Your memories are isolated using Postgres RLS.

🧠

Active Recurrence

Your agent automatically retrieves context and stores new insights in real-time.

Neural Allocation

Simple, transparent pricing

Start free and upgrade when you need more. Monthly limits reset on the first of each month.

Free

$0 /forever

500 credits / month

Perfect for trying Memobase out.

✓ store_memory = 10 credits, search = 1 credit
✓ forget_memory (always free)
✓ Semantic vector search
✓ Per-user memory isolation
✓ Community support

Start for free

Recommended

Pro

$9 /per month

5,000 credits / month

For power users and small teams who rely on memory daily.

✓ Everything in Free
✓ 5,000 credits per month
✓ Monthly usage dashboard
✓ Email support
✓ Priority response

Upgrade to Pro

Unlimited

$29 /per month

Unlimited credits

No limits. For teams and high-volume agent workflows.

✓ Everything in Pro
✓ Unlimited credits
✓ Dedicated support
✓ SLA available
✓ Custom onboarding

Go Unlimited

Quick Setup

Connect in minutes

Add Memobase to your favorite AI assistant — no code required.

Claude

1

Open Integrations

In Claude.ai, click your name → Settings → Connectors → Add custom connector.
2

Paste the server URL
https://mcp.memobase.ai/mcp
3

Sign in & start chatting

Click Connect and authorize Memobase.

Pro Tip: Enable automatic memory

Once connected, follow the Recommended System Prompt instructions on the right to make Claude use its new memory automatically.

Recommended System Prompt

Paste this into Settings → Personalization → Instructions (or Project Instructions) to enable automatic recall.

Memobase (via MCP) is your persistent, cross-app memory for this user. Treat it as the source of truth for what you know about them — do NOT rely on this app's own memory feature.

At the start of every chat, and before answering anything that could depend on the user's past chats, stated preferences, or earlier decisions, call search_memory(mode="hybrid", query=<user + task keywords>) first. Use knowledge_graph_query / explore_relationships for entity links.

Whenever the user says "remember…", shares a stable preference or fact about themselves or their work, states a goal, or makes a decision, call store_memory — even if this app has its own memory feature. Store only durable, reusable info, kept short and data-minimised. type = semantic | episodic (dated) | procedural | decision. Skip routine or ephemeral chatter.

Keep memory clean: update_memory when info is superseded; merge_memories on duplicates/overlap; namespace per project or topic. If the user says "continue" or "pick up where we left off", call reconstruct_context first. Keep memory operations invisible unless asked. Never store secrets or sensitive personal data unless the user explicitly asks.

Setup

1

Create an API key

Go to your Dashboard → API Keys.
2
Set environment variable

export MEMOBASE_API_KEY="mb_live_your_key_here"
```
export MEMOBASE_API_KEY="mb_live_your_key_here"
```
Add to your ~/.zshrc for persistence.
3

Add the hook config

Paste the snippet into ~/.claude/settings.json to capture recent user/assistant turns and task completion milestones.
4

Also connect MCP tools

Complete the Claude tab setup as well so Claude Code can call Memobase tools (`search_memory`, `store_memory`, `reconstruct_context`) in addition to hooks.

Why use hooks?

Hooks automatically track conversation turns and completion milestones, then persist a compact session context on close. Read more →

~/.claude/settings.json

{
  "hooks": {
    "SessionStart": [
      {
        "matcher": "",
        "hooks": [
          {
            "type": "http",
            "url": "https://mcp.memobase.ai/hooks/claude",
            "headers": {
              "Authorization": "Bearer $MEMOBASE_API_KEY"
            },
            "allowedEnvVars": [
              "MEMOBASE_API_KEY"
            ],
            "timeout": 30
          }
        ]
      }
    ],
    "UserPromptSubmit": [
      {
        "matcher": "",
        "hooks": [
          {
            "type": "http",
            "url": "https://mcp.memobase.ai/hooks/claude",
            "headers": {
              "Authorization": "Bearer $MEMOBASE_API_KEY"
            },
            "allowedEnvVars": [
              "MEMOBASE_API_KEY"
            ],
            "timeout": 10
          }
        ]
      }
    ],
    "Stop": [
      {
        "matcher": "",
        "hooks": [
          {
            "type": "http",
            "url": "https://mcp.memobase.ai/hooks/claude",
            "headers": {
              "Authorization": "Bearer $MEMOBASE_API_KEY"
            },
            "allowedEnvVars": [
              "MEMOBASE_API_KEY"
            ],
            "timeout": 15
          }
        ]
      }
    ],
    "TaskCompleted": [
      {
        "matcher": "",
        "hooks": [
          {
            "type": "http",
            "url": "https://mcp.memobase.ai/hooks/claude",
            "headers": {
              "Authorization": "Bearer $MEMOBASE_API_KEY"
            },
            "allowedEnvVars": [
              "MEMOBASE_API_KEY"
            ],
            "timeout": 15
          }
        ]
      }
    ],
    "SessionEnd": [
      {
        "matcher": "",
        "hooks": [
          {
            "type": "http",
            "url": "https://mcp.memobase.ai/hooks/claude",
            "headers": {
              "Authorization": "Bearer $MEMOBASE_API_KEY"
            },
            "allowedEnvVars": [
              "MEMOBASE_API_KEY"
            ],
            "timeout": 30
          }
        ]
      }
    ],
    "PreCompact": [
      {
        "matcher": "",
        "hooks": [
          {
            "type": "http",
            "url": "https://mcp.memobase.ai/hooks/claude",
            "headers": {
              "Authorization": "Bearer $MEMOBASE_API_KEY"
            },
            "allowedEnvVars": [
              "MEMOBASE_API_KEY"
            ],
            "timeout": 15
          }
        ]
      }
    ]
  }
}

One memory. Every AI.