AI product Open source
Remory is an open-source system for adding learned residual memory to large-language-model agent context compaction. It supplements an agent’s generated summary with soft memory conditioned on that summary, allowing a frozen model to continue work after earlier conversation history is discarded.
During compaction, its hook reads the actor model’s hidden states and runs a memory network; during continuation, the saved soft tokens are inserted after the summary. The repository provides SGLang and Transformers deployment, model checkpoints, a Python Harness and Client API, a Codex integration, and training procedures for reconstruction and summary-conditioned residual-memory networks. It is released under the MIT License.
2 uses taken from transcripts — each links to the moment in the video.
Helps AI agents keep working after their context is compacted; it reads the model's hidden states from the discarded conversation history and turns them into a small learned-memory block placed right after the summary, connecting to current agents including Kodak.
An open-source learned residual memory system for context compaction from Fudan's Fusion Lab. A memory network reads model hidden states and stores residual memory in local SQLite, supplementing agent context summaries. Ships Qwen and GLM checkpoints and needs two 80GB GPUs.
2 in the library.