Secondary Mind: From Code with Kiro Hackathon to Production Context Engine
The Journey: Hackathon Submission → Omnia's Intelligence Substrate
The Beginning
Secondary Mind started at the Code with Kiro hackathon tackling a universal developer pain point: developers spend 60% of their time just understanding code before they can make meaningful changes.
The inspiration was personal—the cognitive load of navigating complex codebases, the disconnect between architectural intent and line-by-line implementation. The hackathon solution: create the "long-term memory" for codebases.
The Original Build (Code with Kiro):
- Tauri desktop app (Rust + React)
- Polyglot parsing: TypeScript/JavaScript (swc) + Rust (syn)
- Smart Context Collection: Find usages, dependencies, related files
- Development Profiles: Reusable context packages for AI co-pilots
- AI Context Amplifier: Auto-inject project context into AI prompts
It worked. It solved the problem. The architecture was sound, the parsing was solid, and developers could actually amplify their AI collaboration.
Finding Expanded Purpose
The hackathon solution: Parse TypeScript/Rust codebases, build "Development Profiles" for AI context, amplify co-pilot effectiveness. It worked beautifully as a desktop application.
Then came an opportunity: integrating that proven architecture into Omnia, an agent-native development environment, where it could evolve beyond its original scope.
What It Became
Multi-Language Support (Production Complete)
- 9 Languages: TypeScript/JavaScript, Rust, Python, Go, Java, C, C++, C#, PHP
- 3 Config Formats: JSON, TOML, YAML with schema-aware semantic parsing
- 62 Symbol Types: From TypeScript interfaces to C++ templates to config keys
- Architecture: Pragmatic choices—Rust (syn), TypeScript (swc_core), others (tree-sitter)
Production Hardening
- 90+ Unit Tests: Comprehensive coverage across all parsers
- Error Recovery: Graceful handling of malformed code
- Schema Intelligence: Semantic understanding of package.json, Cargo.toml, GitHub Actions, Docker Compose
- Documentation Extraction: JSDoc, rustdoc, docstrings, JavaDoc, Doxygen, PHPDoc
Real Integration
- Powers Omnia's Context Spaces for library indexing
- Feeds Loom Intelligence Engine for agent memory
- Enables Atlas visualization with symbol nodes
- Supports LQL queries for precise code discovery
The Core Difference
Traditional Approach: Search for "authenticate" → 47 text matches → manual filtering
Secondary Mind: Query "authentication classes" → Parsed symbols with methods, relationships, visibility, documentation
Structured understanding vs text matching.
Status Today
From hackathon prototype to production infrastructure. The question—"can we parse codebases intelligently?"—is now simply: "yes, and here's the architecture that proves it."
Adding new languages? One parser file + three integration points. Schema patterns recognized across config formats. Agent memory built on symbolic understanding.
This is what context-as-infrastructure looks like.
Log in or sign up for Devpost to join the conversation.