Skip to content
All posts

Terminal-First Thinking: Why I Built YourCode

On building a terminal-based coding agent and why I believe the future of developer tools lives in the terminal, not the browser.

Devvrat Hans5 min read
  • #cli
  • #open-source
  • #dx

Every few years, there's a wave of developer tools that try to bring the coding experience into the browser. GitHub Codespaces, StackBlitz, CodeSandbox. They're all great. But there's something about the terminal that these tools can't replicate.

The terminal is the most honest interface a developer has. No animations, no loading spinners, no drag-and-drop abstractions. Just input, output, and the full power of the system underneath.

Why terminal?#

When I started building YourCode, I asked myself: what would a coding agent look like if the terminal was the primary interface? Not a plugin for VS Code. Not a web app. Just a terminal.

The answer turned out to be surprisingly clean:

  • Zero dependencies: no Electron, no browser, no runtime. Just the terminal you already have.
  • Monorepo architecture: CLI, server, database, and shared packages, each independently developable.
  • Persistent sessions: pick up where you left off, even after closing the terminal.
  • Customizable themes: because developers care about how their tools look, even in the terminal.

The architecture#

YourCode is structured as a decoupled monorepo with four packages:

  1. packages/cli: terminal UI powered by OpenTUI and React 19, managing the loopback OAuth listener, keyboard layers, and local tool execution runtime
  2. packages/server: Hono API gateway managing Clerk JWT authentication, Polar usage metering, system prompt compilation, and Vercel AI SDK multi-provider streaming
  3. packages/database: Prisma ORM layer with PostgreSQL
  4. packages/shared: isomorphic Zod validation schemas, model definitions, and tool contracts

Client-Side Distributed Tool Execution#

Most cloud coding assistants force you to upload your entire codebase to remote containers, introducing latency and compliance risks. YourCode decouples model orchestration from tool execution: the backend streams tool call specifications, but every file read, search (grep/glob), edit, and shell command (bash) runs locally on your machine within strict process.cwd() path confinement boundaries. Proprietary source code never leaves your workspace unprompted.

Enterprise AI Governance & Privacy Roadmap#

As agentic coding becomes standard, security and compliance are paramount. We're actively building enterprise governance directly into the CLI runtime:

  • Client-Side DLP & PII Redaction: Scanning tokens for private keys, secrets, and personal information before LLM transmission using Shannon entropy and local NER pseudonymization.
  • Indirect Prompt Injection Defense: Structured XML/Markdown boundary framing and schema AST validation to neutralize malicious instructions embedded in third-party source files.
  • Zero-Knowledge Conversation Encryption: Client-side AES-256-GCM message encryption derived via Argon2id from user passphrases, ensuring even database breaches expose zero plaintext code discussions.
  • Process Sandboxing: Containerized process jailing for arbitrary shell commands with interactive Human-In-The-Loop (HITL) approval gates.

Open source#

I decided to make YourCode open-source from day one. Not because I expect a huge community (yet), but because I believe developer tools should be transparent. You should be able to read the code, understand the decisions, and fork it if my choices don't match yours.

Check it out at yourcode.space or on GitHub.

thanks for reading