by BennettSchwartz · Codex Skill · ★ 82
Membrane Membrane is a selective learning and memory substrate for LLM agents. It gives agents typed, revisable memory instead of an append-only transcript or a flat vector store. Use it when an agent needs to remember facts, retrieve prior context, revise stale knowledge, preserve task state, and cite the records that informed an answer. What It Provides Five memory layers plus entities: episodic, working, semantic, competence, plan graph, and canonical entity nodes that connect them. Entity-connected retrieval: returns ranked records plus bounded graph neighborhoods, not only top-k chunks.
| Stars | 82 |
| Forks | 8 |
| Language | Go |
| Category | Codex Skill |
| License | MIT |
| Quality Score | 71.4893625489717/100 |
| Open Issues | 7 |
| Last Updated | 2026-05-02 |
| Created | 2026-01-31 |
| Platforms | go |
| Est. Tokens | ~104k |
These tools work well together with membrane for enhanced workflows:
Explore other popular codex skill tools:
membrane is A selective learning and memory substrate for agentic systems — typed, revisable, decayable memory with competence learning and trust-aware retrieval.. It is categorized as a Codex Skill with 82 GitHub stars.
membrane is primarily written in Go. It covers topics such as agent, agent-framework, agent-memory.
You can find installation instructions and usage details in the membrane GitHub repository at github.com/BennettSchwartz/membrane. The project has 82 stars and 8 forks, indicating an active community.
membrane is released under the MIT license, making it free to use and modify according to the license terms.