Phoenix V2: A Cognitive Architecture for AI That Remembers, Feels and Evolves — Paper, Code and Book

Description and use

Phoenix V2 is a local AI architecture that stores memory and emotional context in SQLite, enabling an assistant to retain user relationships across model updates.

Problem

AI assistants lose context when the model is updated, causing loss of persistent memory and relationship with the user.

Outcome

Phoenix V2 enables a persistent, emotionally‑aware AI assistant that survives model updates, runs on legacy hardware, and keeps all user data locally.

Workflow

• 1. Store user messages and retrieved memories in a local SQLite database.
• 2. Retrieve relevant memories using semantic similarity, recency, and importance scoring.
• 3. Build a structured prompt with current message, memories, emotional state, and self‑model.
• 4. Send prompt to the LLM for reasoning (single API call).
• 5. Apply local heuristic to draft response; if flagged, revise via LLM.
• 6. Modulate tone with Personality Agent based on affective state.
• 7. Periodically run SubconsciousEngine to consolidate memories and update beliefs.
• 8. Persist all changes back to SQLite; only text of messages and embeddings are sent to external APIs.

Evidence

All persistent state — every memory, every emotional context, the system's model of who you are — lives outside the LLM, in a local SQLite database on your machine.

Source

The entire pipeline requires no GPU. Embedding generation uses a remote API call. LLM calls go to whichever provider you configure — the client is a single file, and swapping providers is a one-line change.

Source