diff --git a/README.md b/README.md
index 2f65499..12ad340 100644
--- a/README.md
+++ b/README.md
@@ -4,8 +4,8 @@
 
 Icebug is a standardized graph format designed for efficient graph data interchange. It comes in two formats:
 
-- **icebug-disk**: Parquet-based format for object storage
-- **icebug-memory**: Apache Arrow-based format for in-memory processing
+- **icebug-disk**: CSR adjacency list/join indices for disk storage
+- **icebug-memory**: CSR adjacency list/join indices for in-memory processing
 
 This project provides tools to convert graph data from simple DuckDB databases or Parquet files containing `nodes_*` and `edges_*` tables, along with a `schema.cypher` file, into standardized graph formats for efficient processing.
 
diff --git a/format/icebug-disk/architecture_overview.svg b/format/icebug-disk/architecture_overview.svg
new file mode 100644
index 0000000..169232c
--- /dev/null
+++ b/format/icebug-disk/architecture_overview.svg
@@ -0,0 +1,36 @@
+<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 820 300" width="100%" height="auto">
+  <defs>
+    <style>
+      .box { fill: #ebf8ff; stroke: #3182ce; stroke-width: 2; rx: 8; ry: 8; }
+      .box-green { fill: #f0fff4; stroke: #38a169; stroke-width: 2; rx: 8; ry: 8; }
+      .box-purple { fill: #faf5ff; stroke: #805ad5; stroke-width: 2; rx: 8; ry: 8; }
+      .text-title { font-family: 'Segoe UI', Helvetica, sans-serif; font-weight: bold; font-size: 14px; fill: #2b6cb0; }
+      .text-green-title { font-family: 'Segoe UI', Helvetica, sans-serif; font-weight: bold; font-size: 14px; fill: #276749; }
+      .text-purple-title { font-family: 'Segoe UI', Helvetica, sans-serif; font-weight: bold; font-size: 14px; fill: #553c9a; }
+    </style>
+  </defs>
+
+  <rect width="100%" height="100%" fill="#ffffff" rx="12" ry="12"/>
+
+  <!-- Components -->
+  <rect class="box" x="10" y="110" width="160" height="80" />
+  <rect class="box" x="15" y="115" width="160" height="80" />
+  <rect class="box" x="20" y="120" width="160" height="80" />
+  <text class="text-title" x="100" y="160" text-anchor="middle" dominant-baseline="middle">Initialization Files</text>
+
+  <rect class="box-green" x="210" y="110" width="160" height="80" />
+  <rect class="box-green" x="215" y="115" width="160" height="80" />
+  <rect class="box-green" x="220" y="120" width="160" height="80" />
+  <text class="text-green-title" x="300" y="160" text-anchor="middle" dominant-baseline="middle">Node Tables</text>
+
+  <rect class="box-purple" x="410" y="110" width="160" height="80" />
+  <rect class="box-purple" x="415" y="115" width="160" height="80" />
+  <rect class="box-purple" x="420" y="120" width="160" height="80" />
+  <text class="text-purple-title" x="500" y="160" text-anchor="middle" dominant-baseline="middle">Indices Files</text>
+
+  <rect class="box-purple" x="610" y="110" width="160" height="80" />
+  <rect class="box-purple" x="615" y="115" width="160" height="80" />
+  <rect class="box-purple" x="620" y="120" width="160" height="80" />
+  <text class="text-purple-title" x="700" y="160" text-anchor="middle" dominant-baseline="middle">Indptr Files</text>
+
+</svg>
diff --git a/format/icebug-disk/spec.md b/format/icebug-disk/spec.md
new file mode 100644
index 0000000..6b2037d
--- /dev/null
+++ b/format/icebug-disk/spec.md
@@ -0,0 +1,56 @@
+# Icebug-Disk graph specification
+
+The Icebug-Disk graph format is designed to store graph data on disk efficiently, enabling high read performance for large graphs.
+
+## Versioning
+
+### Version 1
+
+Initial version of the format. Uses compressed sparse row-based (CSR) adjacency list/join indices for efficient graph traversals.
+
+## Architecture Diagram
+
+![Icebug-Disk Architecture Overview](architecture_overview.svg)
+
+## Components
+
+### Initialization files
+
+Initialization files could be format version, schema definitions, table creation statements etc. These files are used to initialize the graph
+
+### Node tables
+
+Node tables store the actual node data in columnar formats like Parquet, Lance, Vortex, etc. Ideally, one file per table should be sufficient for storing the node data. However, for very large graphs, multiple files can be used to store partitions of the node data.
+
+Example node table:
+
+| id | name  | age |
+| -- | ----- | --- |
+| 1  | Alice | 30  |
+| 2  | Bob   | 25  |
+
+### Indices files
+
+Each indices file[^1] is a table that stores relationship data (i.e., `from_node_id`, `to_node_id`, and edge properties) in columnar format. Similar to node tables, there can be multiple files to store partitions of the relationship data for very large graphs.
+
+Example indices file:
+
+| from | to | weight |
+| ---- | -- | ------ |
+| 1    | 2  | 0.5    |
+| 2    | 1  | 0.8    |
+
+### Indptr files
+
+Each indptr file[^1] (per rel table) stores the offset range (in the indices file) for each node. These files are used to efficiently retrieve the edges connected to a specific node during graph traversals.
+
+Example indptr file:
+
+| edge_id |
+| ------- |
+| 0       |
+| 1       |
+| 2       |
+
+[^1]: https://www.usenix.org/system/files/login/articles/login_winter20_16_kelly.pdf
+
diff --git a/format/icebug-memory/spec.md b/format/icebug-memory/spec.md
new file mode 100644
index 0000000..0d8c400
--- /dev/null
+++ b/format/icebug-memory/spec.md
@@ -0,0 +1,2 @@
+# Icebug-Memory graph specification
+Similar to Icebug-Disk, but entirely in-memory