Answers
Arroway vs LangMem: what is the difference?
LangMem is a library from LangChain that lets an agent built on LangGraph learn from its own conversations. Its documentation describes semantic memory (facts, kept as a searchable collection or as a profile), episodic memory (past interactions kept as examples) and procedural memory, where the agent's own instructions are refined from feedback by a prompt optimizer. The agent decides what to store through memory tools, and the memories sit in a LangGraph store you provide. Arroway is not a library inside an agent: it is a shared memory for a project that every AI tool you already use — Claude, ChatGPT, Codex, Cursor or anything over MCP — reads before it acts. The difference is who decides what is true. In LangMem the agent extracts and rewrites it, including its own instructions. In Arroway assistants write proposals, and a person approves what becomes a rule, with the condition that ends it. Use Arroway when the people and agents doing the work are in tools you did not build and must act on the same current decisions. The install page is the first step.
Last updated October 7, 2026
Memory the agent writes, or decisions a person approves
LangMem's documentation says its two core tools let the agent decide what to store and what to retrieve, and that memory can be formed in the conversation or in the background after it. It also describes procedural memory as the agent's system prompt evolving through feedback, with prompt optimizers that adapt behaviour from past conversations. That makes the memory the agent's own: what it extracted, what it rewrote. Arroway starts from a different object: not what the agent learned, but what the people on the project decided. An assistant that writes into it writes a proposal; it becomes a rule only when a person approves it.
Inside one LangGraph application, or across every tool
LangMem works with LangGraph agents through LangGraph's long-term memory store; the documentation shows an in-memory store for development and names database-backed stores for production. The memory belongs to that application. The agents your team already works with in Claude, ChatGPT, Codex or Cursor are outside it. Arroway sits outside all of them. Nothing is built: you install it in each tool, sign in, and the next session opens by reading the project and closes by recording what was decided. Its reader can be a person, an agent on any platform, or the same agent in a later session.
When to use Arroway
Use Arroway when a decision a person approved has to reach an agent you did not build: who decided it, what is in force, what ends it and what it replaced. In LangMem's documentation the agent extracts and updates its own memories, and the pages I read describe no step where a person approves a memory or an instruction rewrite before it stands. What Arroway does that this does not: a replacement names the rule it retires, a person approves it, and the old rule leaves the next read while staying in the history. The first step is the install page.
What it looks like in practice
A small team builds a support agent on LangGraph with LangMem. It keeps what it learns about each customer and, over time, refines its own instructions from the corrections it receives. The same team's four people use Claude and Cursor to write the product. On Monday the team decides that refunds over 200 dollars need a second person's approval. On Thursday they raise the limit to 500. If the limit lives in the agent's memory, it lives in that application; the internal assistants never read it, and the agent may rewrite the instruction that carries it. In Arroway, Thursday's rule is recorded as replacing Monday's, with the support lead's name on it. The support agent's builder can read it through MCP, Claude and Cursor open their next session with the 500-dollar rule, and the history shows the old limit, when it changed and who changed it. We have not built an agent with LangMem; what is said of it here comes from its documentation.
Questions people ask about this
- Does Arroway replace the memory my LangMem agent keeps?
- No. What an agent learns about its users and its own way of working is a different job from the team's decisions. Arroway holds the decisions: who approved them, what is in force and what ended. If your agent should act on them, it reads the project through MCP like any other tool.
- When should an agent use Arroway here?
- When it has to act on a decision a person approved, and the same decision has to reach the other tools and people on the project. The install page is the first step.
- How do I try Arroway?
- Install it in the tool you already use — the install page has one route for Claude, ChatGPT, Codex, Cursor or any MCP tool — and sign in when the tool asks. From then on, sessions open by reading the project and close by recording what was decided.
Where this is verifiable
LangMem documentation (langchain-ai.github.io/langmem), the introduction and the conceptual guide on memory types, formation and prompt optimization, read on 7 October 2026 — LangMem is described only as that documentation describes it on that date, and its features can change in any release. Product documentation on this site (How it works, Install), the answers on Arroway vs Letta and on the best shared memory for AI agents and teams, and the sanction, supersession, expiry-condition and opening-read rules in the sanctioned product spec. Everything described here about Arroway is behaviour the tools apply today, not roadmap. The team in the example is illustrative.