I noticed Agent Squad already provides conversation-context management and custom storage/retriever seams. The gap I have in mind is a little different from saving a thread: an agent may need a few durable user- or project-approved facts when a new conversation starts, without replaying an old transcript or baking that context into every prompt.
I’m Vivek Gupta, founder of Memcode. Would a small optional integration for Agent Squad make sense—perhaps a retriever plus an explicit save operation exposed through the framework’s existing tool/storage interfaces? It could complement the current conversation persistence, not replace it.
A concrete case: an agency with a research agent and a writing agent can use the same user-authorized project preferences on separate runs, while each agent continues to handle its own conversation history as it does today. The application should choose when memory is enabled and which approved details can be saved; no full chats, system instructions, or tool traces should be uploaded automatically.
For a hosted setup, Memcode offers an MCP endpoint at https://mcp.memcode.in/mcp with OAuth/PKCE and dynamic client registration, so the example would not embed a long-lived API key. Users could revoke access, and normal orchestration should continue if memory is disconnected or temporarily unavailable.
Would you prefer to see this as a retriever example, a separate optional package, or a contribution to a particular runtime first? I can put together a narrowly scoped PR with tests and docs once you point me at the right seam.
cc @brnaba-aws for your thoughts on this integration.
I noticed Agent Squad already provides conversation-context management and custom storage/retriever seams. The gap I have in mind is a little different from saving a thread: an agent may need a few durable user- or project-approved facts when a new conversation starts, without replaying an old transcript or baking that context into every prompt.
I’m Vivek Gupta, founder of Memcode. Would a small optional integration for Agent Squad make sense—perhaps a retriever plus an explicit save operation exposed through the framework’s existing tool/storage interfaces? It could complement the current conversation persistence, not replace it.
A concrete case: an agency with a research agent and a writing agent can use the same user-authorized project preferences on separate runs, while each agent continues to handle its own conversation history as it does today. The application should choose when memory is enabled and which approved details can be saved; no full chats, system instructions, or tool traces should be uploaded automatically.
For a hosted setup, Memcode offers an MCP endpoint at https://mcp.memcode.in/mcp with OAuth/PKCE and dynamic client registration, so the example would not embed a long-lived API key. Users could revoke access, and normal orchestration should continue if memory is disconnected or temporarily unavailable.
Would you prefer to see this as a retriever example, a separate optional package, or a contribution to a particular runtime first? I can put together a narrowly scoped PR with tests and docs once you point me at the right seam.
cc @brnaba-aws for your thoughts on this integration.