Personal-GraphRAG v0.1.0 — first public release

Personal GraphRAG v0.1.0

First public release. A self-hosted knowledge graph for persistent personal
memory that Claude writes to and reads from directly over MCP — no second
extraction model, and the local stack (Ollama) is optional.

Highlights

• Claude is the relationship model: the remember write path is deterministic;
  entities and edges come from structured input, never guessed by a small model.
• Journal-first durability: every write appends to an append-only JSONL journal
  before the graph is touched; rebuild.py reconstructs the graph losslessly.
• No-LLM read path: search_memory and the graph primitives return raw context
  for Claude to reason over.
• Runs with no Ollama: the zero-dependency hash embedding backend runs the
  full pipeline and the entire test suite.

Verified end-to-end (test_memory.py + test_integration.py) on Python 3.10–3.12.
See the README and docs/ARCHITECTURE.md for the design rationale.