Hardening: Full Dynamic Kernel Transition (v0.2.0)

- Removed all static const-generics (D, MAX_RECORDS, MAX_NODES, MAX_EDGES)
- Implemented dynamic heap-allocated pools for Record, Node, and Edge storage
- Zero-Config: Model-agnostic dimension detection and locked-dimension runtime
- Hardened WAL, Event Log, and Replication for dynamic resizing
- V3 Self-describing Snapshot format with backward compatibility
- Precision search and ingestion verified at 10k scale from Python
- Updated README and Docs for the new Zero-Config architecture

Author: varshith-Git

Cargo.lock

@@ -21,6 +21,8 @@ Valori is a vector database built for **regulated industries** (healthcare, fina
 
 **Valori's Solution:** We give you **cryptographic proof**. Bit-identical state hash before and after crash. Zero trust required.
 
+**New in v0.2.0:** **Zero-Config Architecture**. No more hardcoded dimensions or record limits. The kernel adapts to your data on the fly.
+
 ---
 
 ## 🛡️ Crash Recovery: Proven, Not Claimed
@@ -182,8 +184,9 @@ Valori uses **Q16.16 Fixed-Point Arithmetic** instead of standard `f32` floats.
 
 ### 5. Crash Recovery & Durability
 - **WAL & Event Log**: Every operation is synced to disk via length-prefixed logs
-- **Batch Ingestion**: Atomic commits for high-throughput bulk inserts
-- **Snapshots**: Instant checkpointing and restoration
+- **Zero-Config Persistence**: WAL and Snapshots are self-describing, restoring state without manual config
+- **Batch Ingestion**: Atomic commits for high-throughput bulk inserts (10k+ vectors/sec)
+- **Dynamic Snapshots**: Instant checkpointing that scales to millions of records
 
 ### 6. Flexible Deployment
 - **Embedded (Python FFI):** Link directly into Python for microsecond latency

README.md

@@ -22,9 +22,10 @@ This hybrid approach allows AI agents to "remember" in two ways:
 
 ### 1. The Record (Vector)
 The fundamental atomic unit of memory.
-*   **What it is**: A dense fixed-point vector (e.g., 16-dim or 768-dim) representing meaning.
-*   **Storage**: Stored in a contiguous memory pool for O(1) access.
+*   **What it is**: A dense fixed-point vector (e.g., 16-dim or 1536-dim) representing meaning.
+*   **Storage**: Stored in a heap-allocated, dynamic memory pool that grows on demand.
 *   **Addressing**: Identified by a `RecordId` (integer).
+*   **Self-Describing**: The kernel auto-detects vector dimensions from the first ingestion, making it model-agnostic (Zero-Config).
 *   **Metadata**: Optional binary blob (up to 64KB). Deterministically hashed and snapshotted.
 
 ### 2. The Knowlege Graph

docs/core-concepts.md

@@ -27,17 +27,19 @@ maturin develop --release
 
 ---
 
-## 2. Your First Memory (Local Mode)
+# 2. Your First Memory (Local Mode)
+
+Valori is now **Zero-Config**. You don't need to declare vector dimensions or pool sizes upfront—the kernel auto-detects them on the first insertion.
 
 Create a file `memory_test.py`:
 
 ```python
 from valori import ProtocolClient
 
-# 1. Define a dummy embedder (In real apps, use OpenAI/SentenceTransformers)
+# 1. Define an embedder (SentenceTransformers example)
 def my_embed(text):
-    # Returns a 16-dim zero vector for demo
-    return [0.0] * 16
+    # This returns 384 dimensions. Valori adapts automatically.
+    return [0.0] * 384 
 
 # 2. Init Client
 client = ProtocolClient(embed=my_embed)