Elpis is my current flagship project: a terminal coding agent built to continue the work, not restart it.
Long sessions lose shape. Goals get buried in transcripts, decisions disappear after compaction, and a new session often begins with the user explaining the same task again. Elpis keeps the active goal, admitted context, decisions, verification, and next action explicit.
It contains the Codex Rust execution foundation and adds Elpis-owned retrieval, portable workspace state, local memory, provider support, behavior, and interface.
Elpis is under active development. The first public version will be tagged after its release checks pass.
- Voice Commander — local voice transcription with optional refinement, structured output, hotkeys, and automatic paste.
- Skills I Use — operating instructions and small harnesses for coding agents, including first-principles engineering and documentation.
- Rubik's Coding Lab — an interactive 3D environment for inspecting cube state, permutations, and rotation logic.
- Eyes Wide Shut — AI security research for the OpenAI gpt-oss red-teaming challenge.
- Financial Market Analysis — time-series experiments and a two-stage filter/expert pipeline.
- AMR Parsing and Summarization — work on abstract meaning representation and summarization.
- Medical Image Segmentation — computer-vision experiments for segmentation.
Question -> delete -> simplify -> accelerate -> automate
I prefer small systems, explicit boundaries, focused checks, and documentation that says what is implemented and what is not.
- Preparing Elpis for its first public release.
- Studying for a master's degree at IUT.
- Improving agent context, continuity, retrieval, and developer interfaces.




