EverOS solves the core problem that language models forget everything after each turn. Developers who build AI agents repeatedly face the need to store facts, preferences, and procedures across sessions without turning their stack into a heavyweight database cluster. EverOS gives them a lightweight, inspectable memory runtime that writes every piece of information as a plain Markdown file. Those files are the source of truth and can be opened, edited, grepped, or version‑controlled with Git. Under the hood a tiny SQLite table tracks state and queues while LanceDB indexes the same content for fast hybrid retrieval that blends BM25 keyword search, dense vector similarity, and scalar filters in a single query. Because the storage is just files plus two embedded libraries, there is no separate MongoDB, Elasticsearch, or Kafka to manage, which keeps operational cost low for solo developers and small teams.
A second pain point is the lack of procedural memory. Most libraries only keep chat history, forcing agents to relearn how to do the same task each time. EverOS records every completed agent run as a Case. Repeated successful patterns are automatically distilled into reusable Skills that are shared across the agent team, giving the system a self‑evolving memory that improves with use. The workflow is simple: install the Python package, run the local demo to see the loop, add your own API keys, start the FastAPI server, and interact through the OpenAI‑compatible endpoints to add, flush, and search memory. Multimodal ingestion (PDF, image, Office, URL) is available with an optional extra. All of this is released under Apache 2.0, with a managed cloud option that shares the same SDK and format for teams that prefer not to self‑host.
Developers get a memory layer they can see, edit, and extend, without sacrificing speed or scalability.
#AI #Product #OpenSource #LLM #Memory #AgentOps