DEV Community

Cover image for 💎 From Web App to Terminal: How I Ported Traliran AI Hub to C11
Traliran
Traliran

Posted on Originally published at traliran.hashnode.dev

💎 From Web App to Terminal: How I Ported Traliran AI Hub to C11

#ai #showdev #c #cli

I'm the author of Traliran AI Hub. Here's the story of how its terminal version came to be.

Background

I develop Traliran AI Hub — a privacy-first serverless AI client that runs entirely in the browser: API keys live in LocalStorage, requests go straight to the providers, and there are no middleman servers. The concept stuck, but I myself spend most of my time in the terminal — tmux, SSH on remote servers, working without a GUI.

At some point it became clear: I wanted the same functionality, but in the terminal. That's how traliran-ai-hub-cli was born — my own port of Traliran AI Hub written in pure C11 with ncurses. The same feature set, the same privacy, just without the browser (this version may lag behind in updates and features due to the intricacies of native development).

What's Inside

A native TUI client (traliran-hub) in C11 + ncursesw + libcurl. The JSON parser (cJSON) is vendored locally — from the system you only need a compiler and a couple of dev packages. It builds into a single binary:

make            # builds ./traliran-hub
make install    # installs to /usr/local/bin (override with PREFIX=...)
make clean
Enter fullscreen mode Exit fullscreen mode

Run it:

traliran-hub
Enter fullscreen mode Exit fullscreen mode

On first run it creates ~/.cache/traliran-cache/ (config/notes/sessions in storage.json) and a starter IDE workspace.

Quick start:

  1. Press 5 (Settings) → set your Provider and API Key, then Fetch Models and pick a model.
  2. Return to Chat (1) and type a message.
  3. Press ? in-app for the full keymap.

Features I Ported from the Web Version

  • Multi-Provider Support: Groq, Gemini, OpenAI, OpenRouter, DeepSeek, Qwen, GLM, Claude, plus local backends (Ollama / Llama.cpp).
  • ⚡ Multi-Model Compare: pick several models at once — the app fires parallel requests and renders side-by-side cards. Perfect for benchmarking without ten open tabs.
  • 💡 Thinking Mode: reasoning models like DeepSeek-R1 — their chain of thought collapses into a clean dropdown.
  • 👥 AI Group Debate: personas (Optimist, Critic, Technologist) dissect an idea over multiple rounds. My "brainstorming" — solo.
  • ⚡ AI IDE: file manager, editor, snapshot versioning (commit/revert), a persistent coding agent in the right panel, project export to .tar.gz.
  • 🧠 RAG: drop .md/.txt into a knowledge base and the chat retrieves relevant content via a tool protocol. Notes export there and back.
  • 💻 Sandbox: preview generated HTML/JS/CSS right inside the chat.
  • 🏪 Assistant Store: ready-made presets (translator, editor, idea generator).
  • ⚙️ Fine-Grained Tuning: Temperature, Top P, and Max Tokens on the fly.
  • 🌍 Full UTF-8: Cyrillic and other non-ASCII everywhere with proper cursor handling.
  • Vim Mode: F12 toggles vim-style navigation.

A small but important detail: plain hotkeys (q, ?, 1..6, Tab, IDE x) only fire after 3 rapid presses within a second, so typing never triggers them by accident. F-keys and Ctrl combos fire immediately.

Why This Is About Privacy

  • Zero middleman: 100% client-side application, no servers.
  • Direct routing: tokens live only in the local config (~/.cache/traliran-cache/hub-config.json) and go straight to the providers.
  • AGPLv3: freely audit, host, and modify.

As the author, I care that users fully control their data — in both versions there's no "black box" running server-side.

Who It's For

  • Those who live in the terminal and don't want to switch to a browser.
  • Those working on remote/headless servers over SSH.
  • Those who want to compare models across providers in parallel and privately.
  • Those looking for a lightweight alternative to heavy browser-based AI IDEs.

Links

Top comments (0)