whereismyai
NamiDSH
NamiDSH is an open source TypeScript project implementing a miniature version of DeepSeek Harness to study plugin-based agent architectures.
NamiDSH is a miniature TypeScript project that mirrors the architecture of DeepSeek Harness in about one thousand lines of code. It contains a mini version of Cordis with dependency injection and hot-swapping, an 11-package mini version of DSH, and a web-based agent that runs in the browser. Users can connect it to DeepSeek or run it offline using a mock model without an API key. It features zero runtime dependencies on Node.js 22.18 and runs TypeScript directly. The project helps developers understand industrial-grade agent frameworks and plugin systems.
NamiDSH at a glance
- Best for: Software engineer
- Runs on: Web, Node.js, self-hosted, open source
- Pricing: Free
- Released: November 2026
- Category: AI agent tool
- Website: github.com/aa1143/namidsh
Key features
- Mini Cordis with dependency injection and hot-swapping
- Eleven packages matching DeepSeek Harness
- Web-based agent interface for browser dialogue
- Offline mock model support without an API key
- DeepSeek model integration support
- Zero runtime TypeScript execution on Node.js
- Fourteen compatibility test scenarios
Use it for
- Studying plugin-based AI agent architectures
- Learning the internal structure of DeepSeek Harness
- Testing agent loops and dependency injection locally
- Running a lightweight agent web server offline
Alternatives to NamiDSH
- DeepSeek Bot — Enables parallel communication and multi-agent group chats.
- Agensh — Scales to 1,024 self-organizing agents without a central orchestrator.
- Composio — Equips AI agents with tools, managed user auth, and sandboxes across 1,500+ apps
- DSH Desktop — It offers a graphical interface for agent management and built-in remote access.
- DSH Android — Runs directly on Android without root or Termux to automate tasks.
- VoltAgent — Offers open-source flexibility with built-in supervisor agent orchestration and RAG.
- Lightpanda — Lower memory footprint and faster execution than Chrome for automated workflows.
- Sandcastle — Orchestrates AI coding agents in isolated Docker, Podman, or Vercel sandboxes.
- Qq Bridge — Integrates QQ messaging directly with DeepSeek agent sessions and local tools.
- Elephant-Goldfish Model — Uses deep-context agents to draft specs and zero-context subagents to review them.
- Dsh Anchored Standard — It separates initial trajectory selection from later tool use for better performance.
- Harness2 — It recursively refines agent execution structures using parallel and sequential search.
All NamiDSH alternatives