Evidence Before Generation
Retrieves the relevant code, symbols, dependencies, and recent project facts before an agent starts writing.
Open source graph intelligence
CGA
Context Graph Agent helps AI coding agents use about 90% fewer prompt tokens, lower hallucination pressure, and find the right files, symbols, and dependencies faster.
90.44% fewer tokensLive benchmark average across 102 real-code cases.
Lower AI riskReduces hallucination pressure by grounding answers in code evidence.
Faster lookupFind relevant files, symbols, calls, and dependencies faster.
Precise targetsSend focused evidence packs instead of broad file dumps.
Evidence first
The relationship-aware engine behind the speed
Context Graph Agent retrieves the right evidence before generation, so AI coding agents query repository relationships, dependencies, files, and symbols instead of relying on keyword search alone.
Repository Relationships
Lets models query calls, imports, definitions, ownership, and dependency paths, not only keyword matches.
Optimized Retrieval Flow
Moves from impact graph to optimized context to minimal code, keeping prompts focused on what can change.
Precise Targets
Improves file and symbol targeting with dependency awareness, so edits land closer to the real implementation surface.
One-click local stack
Ship CGA as a desktop-grade open-source package
The Docker Desktop bundle starts CGA, PostgreSQL, FalkorDB, Redis, and backups from a single launcher.
Install Docker Desktop
Download CGA-Docker-Desktop.zip
Double-click start-cga-desktop.cmd
Open http://localhost:18001/adminImpact GraphChange-aware traversal
Optimized ContextBudgeted evidence packs
Minimal CodeSmall targeted excerpts
MCP SSEAgent interface
Open source, ready to inspect
Built for public release
The repository includes license, notices, security guidance, contribution terms, and a customer-ready Docker Desktop package.