Projects

Self-hosted agent runtime on one machine

Conversation is only the entry point — the capability lives in the tool layer, the memory layer, and the scheduler. 95 tools are isolated per role, long-term memory both forgets and distills, and six clients share one backend.

Python / FastAPI / C++ / llama.cpp / React / Kotlin / Swift / Docker

Branch-aware visibility for derived memory in tree-shaped histories

Regenerating, editing an earlier message, or forking from a checkpoint turns a conversation into a tree. Filtering raw messages by branch is easy; summaries and facts derived from them are the hard part. branchmem enforces one rule: a derived memory is visible only if all its sources are on the current branch. It ships with a 30-scenario benchmark.

Python / pytest / BM25 / Mem0 / LLM

LLM residency, image bursts, and CPU backfill on a single GPU

A research prototype: on one GPU an LLM stays resident, image generation arrives in bursts, and CPU work backfills the gaps. The work covers admission control, migration cost, and crash recovery; the experimental results go into a thesis.

Python / C++ / CUDA / PyTorch / FastAPI / WebSocket / llama.cpp