About this integration
Persistent memory for Heym agents
Heym is a source-available AI-native workflow automation platform with a visual canvas, Agent nodes, and native MCP support. It lets you build agent workflows visually and run them on a schedule or via webhook.
The community-built Torqon template plugs Torqon MCP directly into any Heym Agent node. When an agent runs, it can store new facts to Torqon's memory store, retrieve semantically relevant context from previous runs, and send long context windows through Torqon's optimizer - all with a single API key and zero extra config.
The result: workflows that remember. An agent that scraped a competitor's pricing last week can reference it this week. A support agent can recall what it told a user three sessions ago. Memory persists across every run, every session, every workflow.
How to set it up
Five steps, one API key
Create a Torqon account
Sign up at dashboard.torqon.dev - the free plan is enough to get started.
Copy your API key
Go to Dashboard ? API Keys and generate a new key. Keep it handy.
Sign up for Heym
Create a workspace at heym.run. Heym is source-available and free to self-host.
Import the template
Click "Use this template" on the Torqon MCP Memory Loop template. It imports into your workspace with the MCP node pre-configured.
Add your API key and run
Paste your Torqon API key into the MCP settings field. That's it - memories are stored and retrieved automatically on every run.
MCP configuration
Paste this into your MCP config
If you're self-hosting Heym or wiring the Torqon MCP manually, use this config block. Replace your-api-key-here with your key from the dashboard.
{
"mcpServers": {
"torqon": {
"command": "npx",
"args": ["-y", "torqon-mcp"],
"env": {
"TORQON_API_KEY": "your-api-key-here"
}
}
}
}
What this unlocks
Memory tools your agents can call
Once connected, your Heym Agent nodes have access to three Torqon MCP tools:
store_memory - Persist any fact, summary, or output from the current run. Stored with a timestamp and searchable by meaning, not exact keywords.
retrieve_context - Semantic search across everything you've stored. Ask "what do we know about user pricing objections?" and get the most relevant stored facts back, ranked by relevance.
optimize_context - Send a long context window and get back a compressed version that preserves the critical signal. Cuts token costs on LLM calls inside your workflows.