Skip to content

semantic-graph-workflow.md

Latest commit

 

History

History
140 lines (97 loc) · 6.57 KB

semantic-graph-workflow.md

File metadata and controls

140 lines (97 loc) · 6.57 KB

Workflow: Building a Knowledge Graph with the Semantic Graph Tab

Overview

The Semantic Graph workflow builds a searchable knowledge graph of the binary's functions, call relationships, security characteristics, and community structure. Once built, the graph enables semantic search, visual exploration, and automated security analysis.

When to Use the Semantic Graph Tab

  • You're starting analysis of a new binary and want a high-level overview
  • You need to find functions related to a specific behavior (network, crypto, file I/O)
  • You want to visualize the call graph around a function of interest
  • You're hunting for vulnerabilities across the entire binary
  • You want to identify functional modules and their relationships
  • You're preparing to share analysis data via SymGraph

Step-by-Step Workflow

Step 1: Open the Semantic Graph Tab

Click the Semantic Graph tab in IDAssist. The bottom status bar shows the current graph statistics. If this is a new binary, the graph will be empty.

Step 2: ReIndex the Binary

Click ReIndex Binary to extract the binary's structure:

  • All functions are indexed via idautils.Functions()
  • Call relationships are extracted from cross-references
  • Data flow edges are identified
  • The progress bar shows indexing progress

This step does not require an LLM — it uses IDA's built-in analysis data. The graph status bar updates to show the number of nodes and edges.

Step 3: Run Semantic Analysis

Click Semantic Analysis to generate LLM summaries for each function:

  • Each function's pseudocode or disassembly is sent to the LLM
  • The LLM generates a natural-language summary
  • Summaries are stored in the analysis database
  • The progress bar shows how many functions have been analyzed

This step uses your active LLM provider and can take a while for large binaries. Options:

  • RAG checkbox: Include document context in the analysis prompts
  • MCP checkbox: Allow the LLM to use tools during analysis
  • Force re-analysis: Regenerate summaries for functions that already have one

Step 4: Run Security Analysis

In the Manual Analysis panel, click Security Analysis to detect vulnerability patterns:

  • Scans for dangerous API usage (buffer operations, format strings, etc.)
  • Identifies security-relevant flags on each function
  • Detects potential vulnerability patterns
  • Results are stored as security flags on graph nodes

Step 5: Run Network Flow Analysis

Click Network Flow to trace network operations across the binary:

  • Identifies functions that send or receive network data
  • Traces data flow from input to output
  • Creates network_send and network_recv edges in the graph
  • Helps understand the binary's network communication patterns

Step 6: Run Community Detection

Click Community Detection to group related functions into logical modules:

  • Uses graph algorithms to identify clusters of related functions
  • Groups functions that call each other frequently
  • Helps identify functional boundaries (e.g., "crypto module", "network handler", "parser")
  • Module information is stored in the graph and visible in search results

Exploring the Graph

List View

The default view showing details for the selected function:

  1. Navigate to a function in IDA (or type a name/address in the Current field and click Go)
  2. Callers and Callees lists show the function's relationships — double-click any entry to navigate
  3. Edges table shows all edges with type filtering — useful for finding vulnerability or taint flow edges
  4. Security Flags section shows and lets you toggle flags on the current function
  5. LLM Summary displays the semantic analysis result — click Edit to modify

Visual Graph

Switch to the Visual Graph sub-tab for an interactive diagram:

  1. The center node (teal) is the currently selected function
  2. Adjust N-Hops (1–5) to expand the visible neighborhood
  3. Toggle edge types: CALLS (blue), VULN (red), NETWORK (gray)
  4. Use zoom controls or scroll to zoom in/out, click Fit to fit the graph to the view
  5. Double-click any node to navigate to that function in IDA
  6. Click a node to see its summary in the panel below

Node colors indicate status:

  • Teal (#2ea8b3) — Currently selected function
  • Dark gray (#3a3f44) — Normal function
  • Dark red (#7a2b2b) — Function with security vulnerability flags

Search

The Search sub-tab provides seven query types:

Query Type Example Use
Semantic Search "functions that parse network packets"
Get Analysis Retrieve stored analysis for a specific address
Similar Functions Find functions similar to the current one
Call Context Explore the call chain around a function
Security Analysis "find all functions with buffer overflow risk"
Module Summary Summarize a detected community/module
Activity Analysis "which functions perform file I/O"

Click a result row to see full details, then click Go To to navigate to that function.

Using the Graph in Queries

The semantic graph enhances other IDAssist features:

  • MCP Integration: When MCP is enabled on the Query or Explain tabs, the LLM can query the graph for function summaries, relationships, and security data
  • RAG + Graph: Combine document context with graph knowledge for the most comprehensive analysis
  • ReAct Agent: The autonomous agent can traverse the graph to investigate relationships across the binary

Tips

  • Start with ReIndex: Always ReIndex before running other analyses — it provides the structural foundation
  • Semantic Analysis takes time: For large binaries (1000+ functions), consider running it overnight or on a subset
  • Use Force re-analysis sparingly: Only re-analyze when you've changed the LLM provider or want better summaries
  • Security flags are cumulative: Running security analysis multiple times won't duplicate flags
  • Refresh Names after renaming: If you rename functions (via Actions tab or manually), click Refresh Names to update graph labels
  • Graphviz improves layout: Install Graphviz (dot command) for better visual graph layouts. Without it, IDAssist falls back to a BFS-based layout.

Related Documentation