01 / Agent runtime
term2
An open terminal agent runtime for local and SSH workflows, with provider-flexible models, guarded execution, durable conversations, and delegated work.
First-party runtime · 5 modes · SSH · Sandbox · Background agents
Duc Nguyen · Software engineer
I build the systems around intelligent software.
I work on agent runtimes, code intelligence, and developer infrastructure—the layers that give AI useful context, safe ways to act, and a place in real workflows.
System trajectory 2012—now
- Applied softwareproducts
- Machine learninginference
- Code intelligencecontext
- AI applicationsinterfaces
- Agent infrastructureruntimes
Moving down the stack—from using intelligent systems to designing how they reason, act, and stay useful.
Selected systems / 01—04
Making developers and AI systems more capable.
Four projects across the stack: runtime, application, context, and inference.
02 / AI application
ChatForge
A full-stack AI workspace for tool orchestration, model comparison, conversation branching, memory, and browser automation.
Next.js · Node.js · SQLite · Electron
03 / Code intelligence
code-rag
Semantic code search that turns syntax-aware code chunks into useful context for developers and coding agents.
Tree-sitter · Embeddings · Vector search · MCP
04 / Applied machine learning
easyocr.js
A TypeScript implementation of the EasyOCR pipeline designed to run in Node.js and browsers without Python at runtime.
TypeScript · ONNX Runtime · Node.js · Browser
Working method
Complex systems, clear boundaries.
- 01
Understand the system
Read the real path before changing the abstraction.
- 02
Find the useful seam
Put complexity behind the smallest honest interface.
- 03
Build across layers
Move from model and runtime through backend and interface.
- 04
Verify behavior
Test the workflow that matters, not just its implementation.
Recent field notes
Writing from the work—and outside it.
Open channel