Support custom `store` for the default ReAct Agent & Deep Agent flows

[anselm94]

Is your feature request related to a problem? Please describe.

Currently, @cap-js/agents wires both ReAct agents (via createAgent) and Markdown-based deep agents (via createDeepAgent) with CdsCheckpointSaver for conversational persistence, but there is no way to attach a LangChain/LangGraph memory store. This means agents cannot persist long-term memory use cases (user preferences, accumulated knowledge, learned skills) across threads, users, or sessions.

For example, an agent cannot "remember that Alice prefers detailed technical explanations" and apply it in a new conversation — every new thread starts from a blank slate.

---

Describe the solution you'd like

Expose store configuration as a first-class extension point in @cap-js/agents, so applications can add a memory store without replacing the full graph construction.

Proposed extension: buildStore event

This should apply consistently to both:

• Deep Agents via createDeepAgent({ store }) - in combination with Support custom backend mounts for the default Deep Agent flows #78 proposal for backend customization to use custom StoreBackend with a tenant/user/agent-scoped namespace, enabling durable cross-thread persistence

• ReAct agents via createFilesystemMiddleware({ store }) - for tools to read/write long-term memory

By default, no memory store is attached, and agents start with a blank slate for each new conversation. Applications may supply custom stores via the buildStore event.

---

Describe alternatives you've considered

1. Override buildGraph completely and construct the LangChain/deepagents graph manually
   This works by copying the current createAutoDeepAgent or createAgent implementation flows, but it is much heavier than overriding a single concern and bypasses the plugin's ergonomic default wiring.

---

Additional context

• LangChain Long-Term Memory
• Deep Agents Memory

---

Have you already checked existing issues before creating a feature request?

Yes, no existing issues cover this feature request.

---

Customer Info

None. CAP Fanboy :)

[github-actions]

👋 Hello @anselm94, thank you for submitting this issue. Our team is reviewing your report and will follow up with you as soon as possible.

[Akatuoro]

Hi @anselm94 , thanks for the feedback! For long-term memory, there are quite a few options. One of them is file-based which is potentially related your other request #78 ? Another option in langchain are the langgraph stores you linked.
CAP has a bit of a leg up compared to raw langchain since we already have access to a persistence and also a way to deploy the necessary schema. So the only remaining part for long-term memory is providing a tool to read / write memories, which you can already do.

[anselm94]

Hi @Akatuoro, true. File-based approach will be perfect for 'markdown-based agents' approach (via #78).

But having a CdsMemoryStore (akin to plugin's CdsCheckpointer) implementation would be beneficial for StoreBackend implementation which uses an underlying memory store for markdown-based agents aka deepagents for scoped memory access when designing /memory mounts for user-specific/session-specific files. While also serving as custom base for building 'long-term memory' for custom LangChain-based/ReAct agents.

P.S: I was building a building a CDS plugin to enable 'LangGraph's persistence layer' to build durable agents. I began with CdsCheckpointer implementation for short-term memory before @cap-js/agents plugin was released to the public. Naturally, CdsMemoryStore is the long-term memory aspect for LangGraph/LangChain/Deep Agents, and have done an implementation already here (I kept it on hold after seeing this plugin getting released). Now that CAP having an official implementation for Checkpointer, I wanted to contribute the second-half of the persistence layer to make it complete, as an ex-SAP employee and a CAP fan ;)