docs: update architecture and api references for cognitive memory

Update AGENTS.md with four-crate architecture (adding llmem-index and llmem-quant), mention Python training pipeline, and update commit guidelines. Update llms.txt project summary with cognitive CLI metaphors and new crate descriptions. Update SKILL.md CLI examples to use new cognitive commands.

Author: urmzd

AGENTS.md

@@ -2,14 +2,18 @@
 
 ## Project Overview
 
-A specification and Rust implementation defining a convention for storing AI agent memory as plain markdown files. Two levels: project (`~/.llmem/{project}/`) and global (`~/.llmem/global/`).
+A specification and Rust implementation defining a convention for storing AI agent memory as plain markdown files. Two levels: project (`~/.llmem/{project}/`) and global (`~/.llmem/global/`). Uses cognitive metaphors for its CLI: memorize, remember, note, learn, consolidate, reflect, forget.
 
 ## Architecture
 
-Rust workspace with three crates:
-- `llmem-core` — spec types, file parsing, index operations
-- `llmem-cli` — CLI binary (`llmem`) for managing memory
+Rust workspace with four crates:
+- `llmem-core` — spec types, file parsing, index operations, inbox, config, embeddings
+- `llmem-cli` — CLI binary (`llmem`) with cognitive commands
 - `llmem-server` — RAG HTTP server for semantic search
+- `llmem-index` — ANN indices (HNSW, IVF-Flat) and tree-sitter code indexing
+- `llmem-quant` — TurboQuant vector quantization (1-4 bit)
+
+Training pipeline in `training/` (Python): data generation, model distillation, ONNX export.
 
 ## Discovering Structure
 
@@ -33,7 +37,7 @@ cargo run -p llmem-cli   # run the CLI
 ## Commit Guidelines
 
 Conventional commits via `sr commit`:
-- `feat(core):` / `feat(cli):` / `feat(server):` — scoped by crate
+- `feat(core):` / `feat(cli):` / `feat(server):` / `feat(quant):` — scoped by crate
 - `docs(spec):` — specification changes
 - `docs(readme):` — documentation changes
 

llms.txt

@@ -1,18 +1,20 @@
 # llmem
 
-> An open standard for tool-agnostic, project-scoped AI agent memory. Defines a two-level memory convention — project (`~/.llmem/{project}/`) and global (`~/.llmem/global/`) — using typed markdown files with YAML frontmatter. Includes a Rust CLI for managing memory and a RAG HTTP server for semantic search. Works with Claude Code, Codex, Gemini, Copilot, Cursor, or any AI tool.
+> An open standard for tool-agnostic AI agent memory with a cognitive CLI. Defines a two-level memory convention — project (`~/.llmem/{project}/`) and global (`~/.llmem/global/`) — using typed markdown files with YAML frontmatter. CLI commands mirror memory processes: memorize, remember, note, learn, consolidate, reflect, forget. Features working memory inbox, attention scoring, Hebbian reinforcement, consolidation cycles, semantic search (HNSW/IVF-Flat), and TurboQuant vector quantization. Includes a RAG HTTP server. Works with Claude Code, Codex, Gemini, Copilot, Cursor, or any AI tool.
 
 ## Docs
 
-- [README](https://github.com/urmzd/llmem/blob/main/README.md): Quick start, CLI usage, and memory types
+- [README](https://github.com/urmzd/llmem/blob/main/README.md): Quick start, cognitive CLI usage, and memory types
 - [AGENTS.md](https://github.com/urmzd/llmem/blob/main/AGENTS.md): Architecture and development guide
 - [Specification](https://github.com/urmzd/llmem/blob/main/SPECIFICATION.md): Full specification (levels, file format, dynamic loading, integration)
 
 ## API / Key Files
 
 - [SPECIFICATION.md](https://github.com/urmzd/llmem/blob/main/SPECIFICATION.md): The formal llmem specification
-- [crates/llmem-core/](https://github.com/urmzd/llmem/blob/main/crates/llmem-core/): Core types and operations
-- [crates/llmem-cli/](https://github.com/urmzd/llmem/blob/main/crates/llmem-cli/): CLI binary
+- [crates/llmem-core/](https://github.com/urmzd/llmem/blob/main/crates/llmem-core/): Core types, inbox, config, embeddings
+- [crates/llmem-cli/](https://github.com/urmzd/llmem/blob/main/crates/llmem-cli/): Cognitive CLI binary
+- [crates/llmem-index/](https://github.com/urmzd/llmem/blob/main/crates/llmem-index/): ANN indices and tree-sitter code indexing
+- [crates/llmem-quant/](https://github.com/urmzd/llmem/blob/main/crates/llmem-quant/): TurboQuant vector quantization
 - [crates/llmem-server/](https://github.com/urmzd/llmem/blob/main/crates/llmem-server/): RAG HTTP server
 - [skills/llmem/SKILL.md](https://github.com/urmzd/llmem/blob/main/skills/llmem/SKILL.md): Portable agent skill
 

skills/llmem/SKILL.md

@@ -117,10 +117,17 @@ Under 200 lines per level. Loaded at every conversation start.
 
 ## Optional CLI
 
+The CLI uses cognitive metaphors for its commands:
+
 ```bash
 cargo install llmem-cli
-llmem init              # project memory
-llmem init --global     # global memory
-llmem add feedback <name> -d "<description>"
-llmem search "<query>"
+llmem init                                          # project memory
+llmem init --global                                 # global memory
+llmem memorize "prefer Rust for CLI tools" -t feedback  # encode to long-term memory
+llmem note "check async runtime choices"            # quick capture to inbox
+llmem learn .                                       # ingest codebase into inbox
+llmem consolidate                                   # promote inbox → long-term memory
+llmem remember "rust"                               # semantic + text search
+llmem reflect --all                                 # review memories + inbox
+llmem forget feedback_old-approach.md               # deliberately forget
 ```