CostWise vs CodeGraph

CodeGraph constructs massive neo4j-like dependency networks in local memories, incurring massive indexing overheads and context bloating. CostWise utilizes incremental AST scope trees to send only relevant declaration scopes instead of raw file contents.

Feature Breakdown

Feature Capabilities	CostWise	CodeGraph
Local AST Scope Extraction	Yes (Tree-sitter)	Yes (Regex-based)
Incremental Index updates	Yes (Watchdog 8ms)	No (Full re-parse)
Token Context Compression	Yes (scoped AST)	No (Sends full graph nodes)
Sub-second Indexing	Yes	No (Heavy JVM process)

CostWise Advantages

•Extremely lightweight SQLite memory footprints.
•Automated client injectors for Cursor & Claude Code.
•Token budget boundaries preventing out-of-context crashes.

CodeGraph Tradeoffs

•Heavy heap allocation on large codebases.
•Requires full re-indexing when structure schema upgrades.
•No cloud sync options (100% local only).

Structural Comparison

CostWise Architecture

A decentralized, standard SQLite key-value mapping file hashes to AST coordinates.

CodeGraph Architecture

A heavy JVM-based graph engine that requires memory-resident graph databases.

CostWise avoids maintaining active runtime pointer chains. Instead, it relies on fast SQL index searches to retrieve definition locations.

Migration Guide: Moving to CostWise

Switching is quick. CostWise integrates into the same Model Context Protocol slots as CodeGraph.

Setup Instructions

Uninstall CodeGraph client configurations from your IDE.
Run the CostWise universal shell loader command.
Launch the auto-injector tool using `costwise install --all`.

// BEFORE (CodeGraph)

{
  "mcpServers": {
    "codegraph-mcp": {
      "command": "docker",
      "args": ["run", "-i", "codegraph/server"]
    }
  }
}

// AFTER (CostWise)

{
  "mcpServers": {
    "costwise": {
      "command": "costwise",
      "args": ["serve"]
    }
  }
}