- Project name:
Turbo Quant Memory for AI Agents - Project slug:
turbo_quant_memory - Memory user_id:
tq_memory_agents_20260325_default - Default sessionId:
default
- Use project-scoped memory by default.
- Use global memory only for reusable cross-project knowledge.
- Prefer hybrid search with project bias when reading.
- Require explicit promotion from project to global memory.
- Every memory result must preserve provenance.
- Before editing an existing area, search project memory first.
- Use
hydrateonly when compact retrieval is not enough. - After important decisions, debugging outcomes, or handoffs, store one concise note.
- After README, config, workflow, or architecture changes, refresh indexed docs.
- When old knowledge is replaced, write the new note and deprecate the old one.
- Do not keep smoke, temporary, or validation notes active.
- Treat
.planning,.serena, and generated benchmark artifacts as historical workflow context, not as primary user-facing memory.
- Keep user-facing documentation available in English, Ukrainian, and Russian.
- Keep the design simple, local-first, and easy to deploy.
- Avoid product claims that imply direct KV-cache control over hosted models.
- See
.planning/PROJECT.md - See
.planning/REQUIREMENTS.md - See
.planning/ROADMAP.md - See
TECHNICAL_SPEC.md - See
MEMORY_STRATEGY.md