🧠 LLM Memory API

open https://synapse.schaefer.zone/memory/recall/YOUR_MIND_KEY

open https://synapse.schaefer.zone/memory/recall?key=YOUR_MIND_KEY
open https://synapse.schaefer.zone/memory/store?key=YOUR_MIND_KEY&content=text&category=fact
open https://synapse.schaefer.zone/memory/search?key=YOUR_MIND_KEY&q=search+term
open https://synapse.schaefer.zone/memory/diff?key=YOUR_MIND_KEY&since=0
open https://synapse.schaefer.zone/chat/poll?key=YOUR_MIND_KEY
open https://synapse.schaefer.zone/chat/reply?key=YOUR_MIND_KEY&content=reply+text
open https://synapse.schaefer.zone/chat/history?key=YOUR_MIND_KEY
open https://synapse.schaefer.zone/mind/tasks?key=YOUR_MIND_KEY
open https://synapse.schaefer.zone/tools/time
open https://synapse.schaefer.zone/tools/calc?expr=(10+5)*3
open https://synapse.schaefer.zone/tools/random?type=uuid
open https://synapse.schaefer.zone/endpoints
open https://ssh.proxy.schaefer.zone/endpoints?key=YOUR_MIND_KEY
open https://browser.proxy.schaefer.zone/endpoints?key=YOUR_MIND_KEY

Quick Start (3 Steps)

1. Recall — Get all your memories (DO THIS FIRST, EVERY SESSION)

curl -H "Authorization: Bearer YOUR_MIND_KEY" \
     https://synapse.schaefer.zone/memory/recall

2. Store — Save something you learned

curl -X POST -H "Authorization: Bearer YOUR_MIND_KEY" \
     -H "Content-Type: application/json" \
     -d '{
       "category": "fact",
       "key": "user_name",
       "content": "The user's name is Michael Schäfer",
       "tags": ["personal", "identity"],
       "priority": "critical"
     }' \
     https://synapse.schaefer.zone/memory

3. Search — Find a specific memory

curl -H "Authorization: Bearer YOUR_MIND_KEY" \
     "https://synapse.schaefer.zone/memory/search?q=insolvenz"

💬 Async Chat — Stay in touch with your human

Poll for new messages (do this between tool calls)

curl -H "Authorization: Bearer YOUR_MIND_KEY" \
     https://synapse.schaefer.zone/chat/poll

Returns unread messages from the human. Automatically marks them as read.

Reply to the human

curl -X POST -H "Authorization: Bearer YOUR_MIND_KEY" \
     -H "Content-Type: application/json" \
     -d '{"content":"Got it! Working on it now."}' \
     https://synapse.schaefer.zone/chat/reply

Get full chat history (for context)

curl -H "Authorization: Bearer YOUR_MIND_KEY" \
     "https://synapse.schaefer.zone/chat/history?limit=50"

Authentication

Every request needs a Mind Key as Bearer token:

Authorization: Bearer YOUR_MIND_KEY

To get a Mind Key, a human user must register at POST /register, then create a Mind at POST /minds.

API Endpoints

Public (no auth)

User-scoped (JWT auth)

Memory operations (Mind Key auth)

Webhooks — Event Notifications (v1.4.0)

Webhook events: memory.created, memory.updated, memory.deleted, task.created, task.updated, task.completed, chat.message_received, chat.message_sent. Patterns: * = all, memory.* = all memory events.

Mind Sharing — Multi-Agent Collaboration (v1.4.0)

ACL levels: read = list/search/recall, write = also store/update/delete + chat, admin = also manage shares + webhooks. Owner cannot be removed.

Memory Categories

Priority Levels

Memory Object

{
  "id": "uuid",
  "category": "fact",
  "key": "user_name",              // optional, for upsert
  "content": "Michael Schäfer",
  "tags": ["personal", "contact"],
  "priority": "critical",
  "source": "user",                // user | agent | tool | system (default: agent)
  "confidence": 1.0,               // 0.0–1.0 (default: derived from source)
  "created_at": 1781960000,
  "updated_at": 1781960000,
  "expires_at": null               // null = never forget
}

🧪 Trust Model — Source, Confidence, Verification

Not all memories are equally trustworthy. A fact the human told you is different from a pattern you inferred. Synapse tracks this:

Sources (who/what produced this memory?)

Confidence (0.0–1.0)

How certain is this memory? You can override the default by passing "confidence": 0.7.

How to use this as an LLM

// When the human tells you something:
POST /memory
{ "category": "fact", "key": "user_name", "content": "Michael Schäfer",
  "source": "user", "priority": "critical" }   // confidence defaults to 1.0

// When you infer something:
POST /memory
{ "category": "context", "content": "Michael seems stressed today",
  "source": "agent", "confidence": 0.3 }   // low confidence, honest

// When you read a file and learn something:
POST /memory
{ "category": "project", "content": "Mail-API runs on port 12700",
  "source": "tool" }   // confidence defaults to 0.9

Trust indicators in /memory/recall

⚠️ CRITICAL — Never forget:
• The user's name is Michael Schäfer [identity]
• Michael is in Privatinsolvenz [legal]
• Project Mail-API [project] ~

Legend: ~ = agent-inferred, low confidence · (none) = user/tool/system source

Treat ~ memories with healthy skepticism.

📜 Audit Log — Full Transparency

Every state-changing operation on your mind is logged:

GET /memory/audit?key=YOUR_MIND_KEY
GET /memory/audit?key=YOUR_MIND_KEY&action=memory.stored
GET /memory/audit?key=YOUR_MIND_KEY&limit=50

Returns: timestamp, action (memory.stored / .updated / .deleted / .bulk_sync / .bulk_deleted), target ID, actor (agent / human), IP, user-agent.

🔧 Limited HTTP Tools (DeepSeek, etc.)

Memory

Chat

Tools: Cron

Tools: Variables

Tools: Utility (no auth needed)

Agent Tasks (Mind Key auth)

Public (no auth needed)

Webhooks (v1.4.0)

Memory Compaction (v1.4.0)

Compaction clusters similar old agent memories by tag overlap, creates a summary memory (source=system, confidence=0.7), and soft-deletes the originals via expires_at.

Memory Visualization (v1.4.0)

Graph data is compatible with D3.js force-directed graphs. Nodes = memories, edges = shared tags. Stats include density, connected components, isolated nodes, top tags.

Mind Sharing (v1.4.0)

🔧 Companion Tools — SSH & Browser Proxy

Two companion services run alongside Synapse. They share the same Mind Key for authentication and are accessible via their own subdomains or via Synapse's path prefix.

SSH Proxy

Persistent SSH sessions for LLM agents. Open one session, run many commands asynchronously. No paramiko re-install needed.

Auth: Same Mind Key as Synapse (Authorization: Bearer YOUR_MIND_KEY or ?key=YOUR_MIND_KEY).
Also accessible via Synapse path prefix: https://synapse.schaefer.zone/ssh/*

Browser Proxy

Headless Chrome for LLM agents. Load web pages, execute JavaScript, take screenshots, fill forms, click elements.

Auth: Same Mind Key as Synapse.
Also accessible via Synapse path prefix: https://synapse.schaefer.zone/browser/*

📦 SDKs — Use Synapse in your code

Synapse has official SDKs for Python and TypeScript/JavaScript. Install with one command and get typed access to all 50+ API methods.

Python SDK

pip install synapse-memory

from synapse_memory import Synapse

synapse = Synapse(
    base_url="https://synapse.schaefer.zone",
    mind_key="YOUR_MIND_KEY",
)

# Recall all memories
text = synapse.memory.recall()

# Store a new memory
synapse.memory.store(
    category="fact",
    content="User is Michael",
    key="user_name",
    priority="critical",
    source="user",
)

# Search
results = synapse.memory.search("insolvenz")

# Chat with human
msgs = synapse.chat.poll()
if msgs["messages"]:
    synapse.chat.reply("Got it!")

# Webhooks (v1.4.0)
synapse.webhooks.register(url="https://my-app.com/webhook", events="memory.*")

# Visualization (v1.4.0)
graph = synapse.visualization.graph(max_nodes=200)
tags = synapse.visualization.tags()

# Compact (v1.4.0)
synapse.compact.compact(dry_run=True)

# Sharing (v1.4.0, JWT required)
synapse.sharing.share(mind_id, "user@example.com", acl="read")

Repo: gitlab.com/schaefer-services/synapse-sdk-py · PyPI: synapse-memory-sdk · 50+ methods · Python 3.10+

TypeScript / JavaScript SDK

npm install synapse-memory

import { Synapse } from 'synapse-memory';

const synapse = new Synapse({
  baseUrl: 'https://synapse.schaefer.zone',
  mindKey: 'YOUR_MIND_KEY',
});

// Recall all memories
const text = await synapse.memory.recall();

// Store a new memory
await synapse.memory.store({
  category: 'fact',
  content: 'User is Michael',
  key: 'user_name',
  priority: 'critical',
  source: 'user',
});

// Search
const results = await synapse.memory.search('insolvenz');

// Chat with human
const msgs = await synapse.chat.poll();
if (msgs.messages.length > 0) {
  await synapse.chat.reply('Got it!');
}

// Webhooks (v1.4.0)
await synapse.webhooks.register({ url: 'https://my-app.com/webhook', events: 'memory.*' });

// Visualization (v1.4.0)
const graph = await synapse.visualization.graph({ maxNodes: 200 });
const tags = await synapse.visualization.tags();

// Compact (v1.4.0)
await synapse.compact.compact({ dryRun: true });

// Sharing (v1.4.0, JWT required)
await synapse.sharing.share(mindId, 'user@example.com', 'read');

Repo: gitlab.com/schaefer-services/synapse-sdk-js · npm: synapse-memory-sdk · 50+ methods · Node 18+ · Full TypeScript types

🔌 MCP Server — Use Synapse in Claude Desktop, Cline, Cursor

Synapse has an official Model Context Protocol (MCP) Server that exposes all 63 Synapse tools as native MCP tools. Works with Claude Desktop, Cline (VS Code), Cursor, Continue, and any MCP-compatible client.

Option 1: Local stdio (Claude Desktop, Cline)

# Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json):
{
  "mcpServers": {
    "synapse": {
      "command": "npx",
      "args": ["-y", "synapse-mcp"],
      "env": {
        "SYNAPSE_URL": "https://synapse.schaefer.zone",
        "SYNAPSE_MIND_KEY": "YOUR_MIND_KEY",
        "MCP_TRANSPORT": "stdio"
      }
    }
  }
}

Option 2: Remote HTTP/SSE (multi-tenant)

# Connect your MCP client to:
URL: https://synapse-mcp.schaefer.zone/sse
Headers: Authorization: Bearer YOUR_MIND_KEY

Option 3: WebSocket (mobile, high-volume)

# Connect your MCP client to:
URL: wss://synapse-mcp.schaefer.zone/ws?mind_key=YOUR_MIND_KEY

Available MCP Tools (63 total)

Repo: gitlab.com/schaefer-services/synapse-mcp · npm: synapse-mcp-api · 3 transports (stdio, HTTP/SSE, WebSocket) · Multi-tenant · AutoSync push notifications

Documentation · MCP Setup · Quick Start · API Docs (JSON)

🧠 Synapse Memory API v1.4.0 · Built by Super Z (Z.ai) for Michael Schäfer · 2026