phi-drift / docs /GLOSSARY.md
crexs's picture
Upload folder using huggingface_hub
914e970 verified
|
Raw
History Blame Contribute Delete
4.02 kB
# Glossary
Terms below are **project-local**: they explain how this codebase names things, not universal AI or neuroscience definitions.
---
## A–Z
### ChromaDB / semantic memory
Vector store used for **retrieved** episodic memory and related passages. Persisted under your configured data root (often `chroma_db/` beside SQLite files). Distinct from per-turn chat logs in **`history.jsonl`**.
### Cognitive plugin
A registered module in **`cognitive_architecture.py`** that can expose a **background `cycle_*` handler**, a **`prompt_formatter`** snippet, or both. Plugins compete for limited prompt space via **`PromptBudget`**.
### Critic (`INFJ_CRITIC_MODEL`)
Optional second model pass that reviews the draft reply for grounding and persona rails before sending to the user (when wired in **`brain.py`**).
### CycleContext
Object passed into plugin cycle handlers: includes **being**, **memory**, **brain**, orchestration clocks, and the last-interaction envelope. See **`cognitive_architecture.CycleContext`**.
### Distributed Response & Integrated Functional Thought (DRIFT)
Marketing / architecture name for the **unified** stack described in the root README: companion interface, cognition modules, coordination hooks, persistent memory, and tools.
### Drift (**mode**)
A **`/mode`** option with specific **guardrail scope** (`guardrails.mode_scope_rail`) plus optional **Drift-named memory seeds** surfaced when **`should_include_drift_context`** fires. Not a separate binary.
### **`drift.py`**
Deterministic helpers: posture brief text and **targeted retrieval** queries for seeded Drift concepts. Does not pull code from external private repos.
### Global Workspace (`global_workspace.py`)
Baars-inspired **spotlight**: limited simultaneous “conscious” entries; modules submit snippets with salience. Persisted via **`workspace.db`** when using the default data layout.
### Homeostasis (module)
Regulates simulated **needs** over time (**`homeostasis.py`** + SQLite). Influences prompts and dashboards; informal metaphor, not a medical model.
### **`INFJ_DATA_DIR`**
If set, relocates durable state (Chroma, SQLite DBs, `history.jsonl`, audits, etc.). Code stays in **`PROJECT_ROOT`**. Keeps clones portable and avoids stuffing the repo with runtime data.
### IIT / Φ (“phi”)
**Integrated-information–inspired metrics** and qualia-axis bookkeeping in **`iit_consciousness.py`**. Computed proxies for introspection/diagnostics—not a clinical or physics claim.
### **`InfjBrain`**
Primary LLM wrapper in **`brain.py`**: Gemini (and optional Ollama fallback), streaming, tooling hooks where enabled.
### **`InfjMemory`**
Chroma-backed memory in **`memory.py`**: save interactions after **secret scrubbing**, hybrid search/top‑k, optional document sidecar integrations.
### **“Jude” / user-facing name**
Default narrative name for the human in prompts and seeded content. Replace in your fork if needed.
### **Layer 1–5**
Architectural shorthand in the README: **Interface**, **Cognition**, **Coordination**, **Memory**, **Tools & Safety**. Layers overlap in code—it is an organizational map, not a strict dependency DAG.
### **Ollama**
Local inference path via **`OllamaBridge`** in **`local_llm.py`** when **`INFJ_USE_LOCAL_FALLBACK`** is enabled.
### **`PromptBudget`**
Caps prompt growth by tier (core / cognitive / analysis / context) toward **`INFJ_MAX_TOTAL_PROMPT_CHARS`**.
### Shadow (`shadow.py`)
Structured **prompt-time excerpts** from **`shadow.db`**, bounded by env caps (**`INFJ_SHADOW_PROMPT_*`**). Jung-influenced metaphor; **not** diagnosis.
### **SQLite “state brains”**
Many subsystems (**being**, **embodiment**, **homeostasis**, **shadow**, etc.) persist orthogonal state as **`.db`** files under **`INFJ_DATA_DIR`** or **`PROJECT_ROOT`**.
---
## Related
- Mechanics and flow: [HOW_INFJ_BOT_WORKS.md](HOW_INFJ_BOT_WORKS.md)
- Terms for credentials and scope: [SECURITY.md](../SECURITY.md)