Engrammar
AI · KNOWLEDGE
Persistent Memory That Helps Coding Agents Learn
Engrammar gives coding agents a memory that improves through use. It learns useful lessons from corrections, repeated problems, debugging, tooling quirks, and project conventions. When a later task matches, Engrammar brings back the relevant guidance automatically. Feedback helps it keep useful lessons, merge repeated ones, and filter advice that no longer applies. Developers can edit, pin, deprecate, or isolate memories.
- Period
- February 2026 to present. About seven months.
- Company
- Rococode
- Role
- Founder, architect, and engineer. I designed the memory model, retrieval pipeline, agent hooks, MCP interface, and learning workflow. Private project. No public link. Apache 2.0.
- Technologies
- Python, SQLite, fastembed, ONNX, BM25, vector search, reciprocal rank fusion, MCP, Claude Code hooks