memodb-io/acontext is an open source skill memory layer for AI agents. The repository description reads "Agent Skills as a Memory Layer", and the repository's latest release is tagged sdk-ts/v0.1.21. The repository was created on July 16, 2025. The goal of this layer is to accumulate what an agent learns during a run into reusable skill memory.
The core is to capture the learning from a single run automatically and store it as skill files that can be read, edited, and shared. Memory is managed as Markdown files that follow a user-defined SKILL.md structure, and the get_skill and get_skill_file tools pull in what is needed. Python and TypeScript SDKs come together with acontext-cli for self-hosting. The primary language is JavaScript, the license is Apache-2.0, and the star count is 3696.
The README, however, only describes the process as far as run experience being saved to files and read back. No benchmark reproduces accuracy or performance, so it is hard to tell from figures how reusable a given experience is. The SDKs provided are only Python and TypeScript, so other language environments need a hand-built integration. Memory also has to be managed directly under a user-defined SKILL.md structure, so a team needs an established way of organizing skills before this fits in easily. Since it is released under the Apache-2.0 license, the first thing to weigh is whether it matches the team's skill management practice.