Shared memory for smolagents
smolagents is Hugging Face's minimalist agent framework, a small core where agents act by writing and running code, calling tools, and iterating in tight loops. The minimalism is the point, but it also means there is no built-in long-term memory: when an agent finishes, what it learned evaporates unless you build a store and a write path yourself, and each agent you spin up starts from a blank slate. Glen, shared memory for AI agents, gives your smolagents long-term shared memory as a single MCP tool, so your lean agents gain durable, organization-wide recall without dragging in a memory stack of their own.
Add Glen to a smolagents setup over MCP and any agent can call one tool that retrieves relevant long-term context and records new facts in a single round trip. It fits the framework's philosophy: instead of bolting a vector database, an embedding model, and bespoke read and write logic onto a deliberately small agent, you give it one tool that handles memory end to end. A CodeAgent can pull what your organization already knows before it starts writing code, then record what the run discovered when it finishes.
The payoff is sharing and persistence that a single lightweight agent cannot provide on its own. Glen is org-scoped, so the memory is shared across every smolagents run, every agent, and every deployment in your organization. One agent learns a durable fact; the next agent reads it, even an agent running a different model or built on a different framework. Because Glen is a standard MCP server, the same memory is readable from Claude Code, Cursor, or any other MCP client, so the knowledge your agents accumulate is shared with your developers and other tools too. You connect once over OAuth or an API key, keep your agents small, and let the memory compound across every run.
FAQ
- Does adding Glen bloat my otherwise minimal agents?
- No. Glen is one MCP tool, not a memory framework you embed. Your agents stay lean and gain durable, org-shared recall through a single call that both reads and writes.
- Can different smolagents share what they learn?
- Yes. Glen is org-scoped, so every agent in your organization reads and writes the same memory, including agents on other frameworks and MCP clients.