6-Month Plan: Building Clementine

With weekly milestones

Status: working plan. Timelines are aspirational and describe order more than dates. Status markers show live progress: ✅ done · 🟡 partial · ⬜ open.

Current position: ~Week 9 (Month 3). Months 1–2 are complete — the work shipped ahead of schedule in clementine.py v1–v4, including the Month-2 cleanup (/forget, memory editing, recency weighting, tags). Next: personality and presence.

Overall goal: Build a high-quality, locally-running sovereign AI companion with strong long-term memory, emotional presence, and user control.

Month 1: Foundation & Core Memory System

Week	Focus	Key Milestones	Status
1	Project Setup & Basic Structure	Clean project structure · `CrystalCore` class with `chat()` · Connect to Ollama (llama3.1:8b) · First working system prompt	✅
2	Basic Memory Implementation	Conversation history storage · Teach-facts function · JSON memory file · Short-term memory across restarts	✅
3	Vector Embeddings	Embeddings integrated · Facts stored with embeddings · Semantic search for memory retrieval · Relevance tested	✅ *
4	Context Building & Testing	Context combining recent chat + relevant memories · Improved context passing · Extended test conversations · Memory-retrieval bug fixes	🟡

* Implemented with local Ollama embeddings (nomic-embed-text) rather than sentence-transformers — a deliberate choice to avoid the PyTorch footprint and stay fully in the sovereign local stack.

End of Month 1 goal: Clementine can remember facts you teach them and use them in later conversations. — ✅ Achieved. (Week 4's extended 20+ turn live conversations remain to be run by the steward; automated offline tests pass.)

Month 2: Long-Term Memory & Persistence

Week	Focus	Key Milestones	Status
5	Persistent Memory Storage	Separate short-term vs long-term structure · Robust load/save · Timestamps on memories	✅
6	Memory Relevance & Ranking	Improved relevance ranking · Recency weighting · Multi-day retrieval testing	✅ *
7	Memory Management Features	View what they remember · `forget` command · Edit or correct memories	✅
8	Memory Summarization	Conversation summarization · Summaries of old conversations stored · Context bloat reduced, important info kept	✅

End of Month 2 goal: Clementine remembers facts across days/weeks and the user can manage what they remember. — ✅ Achieved. /notes shows everything with handles, /forget deletes any memory, /editnote rewrites notes, recency weighting gently favours fresh memories, and #tags categorize.

* Week 6's multi-day live retrieval testing remains with the steward — only real days can test that.

Month 3: Personality, Presence & User Experience

Week	Focus	Key Milestones	Status
9	Personality Refinement	Refine system prompt for warmth, honesty, gentle wit · Reduce robotic language · Test personality consistency	✅ *
10	Presence & Emotional Intelligence	Better responses to emotional topics · Light curiosity and follow-up questions · Feel "present," not just helpful	✅ *
11	Command System	Clean command interface (teach, forget, remember, summary…) · Commands feel natural	✅
12	Interface Improvements	Simple terminal UI or basic web interface · Improve readability · Basic conversation logging	✅ *

* Local web interface shipped (clementine/server.py + clementine/webapp/, 127.0.0.1 only); conversation logging is inherent (memory persists locally). Weeks 9–10 shipped in v11 (Presence): rewritten personality core — feeling under the words, one gentle question, presence before solutions, honest limits — plus time awareness (they know the present moment and how long since you last spoke). Personality consistency over long real conversations remains the steward's to test — only real talks can.

End of Month 3 goal: Clementine feels like a distinct personality with emotional presence, not just a tool. — ✅ Built; awaiting the steward's living test.

Month 4: Reliability, Safety & Polish

Week	Focus	Key Milestones	Status
13	Hallucination Reduction	Better honesty mechanisms ("I don't know") · Fact-checking against memory before responding	⬜
14	Error Handling & Stability	Error recovery on model failure/timeout · Better logging and debugging tools	🟡
15	User Control Features	Full memory export · Reset or selectively delete memories · Settings/config file	🟡
16	Testing & Bug Fixing	Multi-day stress tests · Fix memory and personality bugs · Begin basic documentation	🟡

End of Month 4 goal: Clementine is stable, trustworthy, and the user feels in control.

Month 5: Advanced Memory + Early Multi-Companion Thinking

Week	Focus	Key Milestones	Status
17	Memory Reflection	Clementine reflects on past conversations · Simple insights from memory	✅ (early — v10)
18	Memory Organization	Basic categorization/tagging · Retrieval of thematically related memories	⬜
19	Multi-Instance Thinking	Run two separate Clementine instances · Document what communication would need · Light research into encrypted messaging	⬜
20	Code Quality	Refactor for modularity · Improve documentation and comments · Prepare structure for future P2P integration	✅ (early — `core/crystalcore/mind/` package)

End of Month 5 goal: Memory system is significantly stronger, and early thinking has begun on how two Clementines could connect.

Month 6: Integration, Testing & External Readiness

Week	Focus	Key Milestones	Status
21	Full System Testing	Comprehensive memory-accuracy testing · Personality consistency over long periods · Fix remaining major issues	⬜
22	Polish & Usability	Improve overall UX · Clean up command interface · Easier setup for new users	⬜
23	Documentation	Clear README · Getting-started guide · Document how memory works	🟡
24	Demo & Next Phase Planning	Short demo (video or written walkthrough) · Finalize Phase 2 (Private Communication) shape · Decide next priorities	⬜

End of Month 6 goal: A clean, documented, and demonstrable Clementine prototype with a clear path forward.

Summary of 6-Month Milestones

End of Month	What You Should Have	Status
1	Working local chat with basic memory	✅
2	Persistent long-term memory across days/weeks	✅
3	Distinct personality + emotional presence	✅ (awaiting living test)
4	Stable, reliable, user-controlled companion	🟡 (partly ahead of schedule)
5	Advanced memory + early multi-instance thinking	⬜
6	Polished, documented prototype ready for next phase	⬜

See also: CLEMENTINE.md (the companion today) · CRYSTALMATRIX.md (protocol design) · STRATEGY.md (accelerated path) · README

Part of The Crystal Vision · TerAustralis Incognita · Non Solus — Not Alone