Torqon MCP Memory Loop

Torqon MCP Memory Loop

Use Torqon as a persistent memory and context optimization layer inside a Heym Agent workflow. This community template keeps the flow explicit: first store a memory, then retrieve relevant context, then optimize the context window before returning a response.

What this workflow does

1. memoryRequest accepts a memory note, a question, and optional extra context
2. torqonMemoryAgent starts the Torqon MCP server locally with npx
3. The Agent calls store_memory with the memory note
4. The Agent calls retrieve_context with the question
5. The Agent calls optimize_context on the retrieved context plus the extra context
6. memorySummary returns the stored-memory summary, retrieved context, and optimized context

Setup

Create a Torqon API key from dashboard.torqon.dev, then import this template into Heym. Open torqonMemoryAgent, find the Torqon MCP connection, and replace <user-api-key> in the TORQON_API_KEY env field with your own key. Heym runs the MCP server with npx -y @torqon/mcp@latest.

Connect an LLM credential to torqonMemoryAgent, run the workflow once with the sample fields, then swap in your own memory notes and questions.

Memory scope

Personal memory is tied to the API key, so only that key can read and write the stored memory. Workspace memory is shared across an organization, so member keys read and write to the same memory pool. The Torqon tools stay the same; scoping happens automatically from the key.

Notes

This first version intentionally uses the explicit store_memory -> retrieve_context -> optimize_context sequence instead of auto_process. The two-step memory loop is easier to inspect, and adding optimization shows where heavy context windows can shrink before the final Agent response.