Hey Product Hunt 👋 Every time my coding agent started a task, it rediscovered my repo from scratch, even for changes it had made a dozen times before. That's slow, and the tokens add up. Singularity gives it a memory. After a session ends with a passing change, it records what happened. After a few of those, it learns workflows like "adding a setting here touches these places and gets tested this way" and the traps to avoid. When a new task starts, your agent gets just the workflows that fit, pointed at today's code. What we measured (medians of 3–4 runs each, hidden tests decide pass/fail): 📉 Long task, all changes familiar: 1.9M → 423k tokens, $0.81 → $0.30, 28 → 10 turns 📉 Half familiar, half new: about 40% cheaper ➖ Brand-new kinds of task: no gain, and no loss ✅ One command to install; memory is stored on your machine ✅ Works in Claude Code, Codex, Gemini CLI and Droid ✅ No always-on instructions: memory shows up only when a task needs it I'd love feedback, especially from big repos where your agent keeps doing the same kinds of change.

Singularity

Memory that makes your coding agent cheaper on repeat work

Memory that makes your coding agent cheaper on repeat work