2.0.0

🚀 Release v2.0.0 — Traliran AI Hub CLI / TUI Edition

First major update of Traliran AI Hub CLI — a lightweight, privacy-first and fully serverless AI terminal client (traliran-hub) written in pure C11 (ncursesw + libcurl).

Version 2.0.0 fixes all shortcomings of the 1.0.0 branch and brings the main feature of this release — MCP (Model Context Protocol) support, fully mirroring the functionality of the Web Edition.

Why 2.0 instead of 1.1? The release was initially planned as 1.1.0, but adding MCP turned out to be a far more significant change than just fixing model comparison, so it was decided to bump the major version. 2.0.0-test-3 is a direct continuation of 1.1.0-test-2, is fully stable, and all of its code will be included in the final 2.0.0.

---

⚡ Quick Install (Linux x86_64)

Option 1: Binary Installation (For rolling-release distros — built with newer library versions and may not run on Debian etc.)

Download the traliran-hub file attached to this release, make it executable and move it to your PATH:

chmod +x traliran-hub
sudo mv traliran-hub /usr/local/bin/

Launch from anywhere:

traliran-hub

Option 2: Build from Source

If you (gentoo user) prefer to build it yourself:

git clone https://github.com/Traliran/traliran-ai-hub-cli.git
cd traliran-ai-hub-cli
make
sudo make install

---

🪟 Note for Windows Users

There is no native Windows .exe build and there never will be. If you are on Windows, ask yourself: why do you even need a Linux terminal client? 😉

If you still want to use the Traliran AI Hub ecosystem on Windows without WSL:

1. Use the web version: Traliran AI Hub Web Edition — runs in any browser.
2. Try Traliran AI Hub DriftWM — a browser-based window manager with infinite desktops and tiling.

Please do not open GitHub Issues complaining that the binary doesn't run in CMD/PowerShell.

---

✨ What's New in 2.0.0

🧩 Highlight — MCP (Model Context Protocol) Support [NEW]

A full MCP implementation, just like in the HTML/CSS/JS version:

• Connect remote MCP servers via the Streamable HTTP protocol
• Servers are configured in the 🧩 MCP Servers modal
• Their tools are automatically forwarded to the model as function calling tools — exactly the same as the IDE agent tools
• The model decides when to call a tool → the hub proxies the call → returns the result to the model → the loop repeats
• Works with any MCP endpoint that supports Streamable HTTP

Creating MCP surprisingly took less time than fixing cross-provider model comparison. Probably because while working on multi-model I had to test and adapt prompt engineering for the new model Muse Spark 1.2 in parallel, which fully wrote all the code you see in this update.

⚡ Cross-Provider Model Comparison — Now Works 100%

In v1.0.0 and 1.1.0-test-1 the feature was broken:

• Only the model selection interface worked
• Due to incorrect saving of multiple API configurations, it was impossible to compare models from different providers, or even just switch between providers

Fully fixed in 2.0.0:

• Parallel requests to multiple models at once
• Mix any providers: Groq + Gemini + OpenAI + OpenRouter + DeepSeek + Qwen + GLM + Claude + Ollama / Llama.cpp
• Side-by-side comparison cards rendering for instant benchmarking
• Synchronization of endpoint ↔ provider ↔ API key for each provider is now correct

---

🐛 Bug Fixes

• Fixed crash when chatting — a crash that occurred when deleting chats has been eliminated.
• Fixed settings synchronization: when switching providers in Settings, the endpoint is now correctly updated, the saved API key is restored and the model list is refreshed. Previously the endpoint got "stuck" on the previous provider.
• Fixed saving of API keys for different providers: each provider now stores its own key and endpoint independently. Switching no longer overwrites data.
• Fixed responses from multiple models of the same provider: now all selected models of the same provider respond correctly (previously only one responded).

🔧 Build & Compatibility Improvements

• Reworked Makefile for better compatibility with different distros and compilers
• Reduced final binary size
• cJSON is still vendored in vendor/cjson/ — no external installation required
• Version in code bumped to 2.0.0, binary rebuilt for release

---

🌟 All Key Features (as of 2.0.0)

• 100% Privacy & Serverless: Direct requests to provider APIs from your machine. No proxies, telemetry, or logging. Keys are stored locally in ~/.cache/traliran-cache/storage.json.
• ⚡ Parallel Model Benchmarking: Compare any models from any providers simultaneously.
• 🧩 MCP Tools: Connect external tools via MCP and extend model capabilities without changing code.
• 💡 Reasoning / Thinking Blocks: Native rendering and collapsible thinking blocks for DeepSeek-R1 and other reasoning models.
• 👥 AI Group Debate: Debate mode with roles Optimist, Critic, Technologist — multiple rounds of cross-analysis of your idea.
• ⚡ AI IDE & Workspace: Full-screen code editor, agent panel, version control (commits/reverts), workspace export to .tar.gz.
• 🧠 RAG Knowledge Base: Upload .md / .txt and chat with documents (retrieval via tools, like in the IDE).
• 💻 Built-in Sandbox: Local preview of generated HTML/JS/CSS.
• 🏪 Assistant Store: Marketplace of system prompts (free and premium).
• ⌨️ Full Vim Support: Toggle anytime with F12.
• 🌍 Full UTF-8 Input: Cyrillic and other characters everywhere — chat, notes, IDE, dialogs.

---

🎹 Current Keymap

Screen	Keys
Global	1..6 switch screens (or Tab / Ctrl+1..6) · ? help · q quit · F12 Vim mode
Chat	Enter send · F2 new session · F3 rename · F4 delete · F5 summarize · F6 regenerate · F7 multi-model · F8 debate · F9 sandbox last code · F10 attach file · PgUp/PgDn scroll
IDE	F1 files / F2 versions / F3 bots · F4 new file · F5 save · F9 commit · F10 revert · x delete · PgUp/PgDn scroll editor
Notes	F2 new note · F3 delete · F4 AI complement · F5 export MD · F6 export to RAG · F7 import MD · F8 preview · F9 search · F10 title
RAG	F2 upload .md/.txt · Tab focus files/chat · Enter chat with documents · F10 clear chat
Store	↑/↓ navigate · Enter install free assistant · o open paid
Settings / MCP	↑/↓ select · Enter edit · Esc back

Hotkeys on plain characters (q, ?, 1..6, Tab, x in IDE) only trigger after 3 rapid presses within 1 second — so they don't interfere with typing. F-keys and Ctrl combos trigger instantly.

---

📦 History of Test Builds Included in 2.0.0

• 1.1.0-test-1 — broken cross-provider comparison, endpoint not updated on provider switch
• 1.1.0-test-2 — fixed endpoint/provider/API keys synchronization, cross-provider comparison works, chat crash fixed, Makefile improved
• 2.0.0-test-3 — fully functional pre-release, all code of which went into 2.0.0. MCP added. Building the C version is significantly harder than the web version, so it was lagging behind, but feature parity has now been restored.

---

Licensed under AGPLv3. Source code and documentation are in the repository.

Download traliran-hub from Assets below and run traliran-hub to get started.

2.0.0-test-3

This pre-release version 2.0.0-test-3 is fully functional; all of this code will be included in the 2.0.0 release.

Why 2.0.0-test-3? Because it's a continuation of 1.1.0-test-2. I just thought the addition of MCP was a more significant development, and it deserved to be called 2.0.

What I want to say is that creating MCP took much less time than I thought, and less than creating the model comparison feature for multiple providers. This is probably because while working on the model comparison feature, I was testing a new AI model, Muse Spark 1.2, and it took time to adapt the prompt engineering technique to this model.

As soon as I prepare a nice Release Note for the 2.0.0 release, I'll immediately merge it with the main branch, change the version in the code to 2.0.0, rebuild the binary, and publish the new release.

1.1.0-test-2

A bug has been fixed with synchronization of endpoints, providers and saving api keys for different providers, so comparing several models from different providers also works properly. Changed Makefile for greater compatibility and less space.
Fixed bug: crashing when chatting.

This version is usable, but not worthy of being a release since MCP must also be added to release 1.1.0

Creating and maintaining a C version is much more difficult than a web version, so it's so far behind.

1.1.0-test-1

I am aware that this release isn't functioning quite right—specifically the multi-model feature (which seemed to have issues in the main release as well), but I am actively working on refining it.
The only part currently working correctly is the multi-model selection interface; however, due to a flawed implementation of how multiple API configurations are saved, it doesn't yet support comparing models across different providers.

There is also an issue in the settings where the endpoint doesn't update when the provider is changed; I plan to fix this by the 1.1.0-test-2 release.
For the final 1.1.0 release, the plan is to add cross-provider model comparison and MCP support—mirroring the functionality found in the HTML/CSS/JS version.

1.0.0 - initial

🚀 Release v1.0.0 — Traliran AI Hub CLI / TUI Edition

First stable release of Traliran AI Hub CLI — a lightweight, serverless, and privacy-first AI terminal client (traliran-hub) written in native C11 (ncursesw + libcurl).

---

⚡ Quick Install (Linux x86_64)

Option 1: Binary Installation (Recommended)

Download the traliran-hub executable attached to this release, make it executable, and move it to your system PATH:

chmod +x traliran-hub
sudo mv traliran-hub /usr/local/bin/

Now you can launch it from anywhere in your terminal:

traliran-hub

Option 2: Build from Source

If you (gentoo user) prefer building it yourself:

git clone https://github.com/Traliran/traliran-ai-hub-cli.git
cd traliran-ai-hub-cli
make
sudo make install

---

🪟 Note for Windows Users

There is no native Windows .exe build available. If you are on Windows, ask yourself: why are you even using a Linux terminal client? 😉

If you still want to use the Traliran AI Hub ecosystem on Windows without WSL:

1. Use the main web version: Traliran AI Hub Web Edition — runs directly in any browser.
2. Check out Traliran AI Hub DriftWM — an in-browser window-manager experience with infinite desktops and tiling modes.

Please do not open GitHub Issues complaining that the binary doesn't run natively on Windows CMD/PowerShell.

---

🌟 Key Features

• 100% Privacy-First & Serverless: Direct requests to API providers (Groq, Gemini, OpenAI, OpenRouter, DeepSeek, Claude, Ollama, Llama.cpp) straight from your machine. No proxy servers, no telemetry.
• ⚡ Parallel Multi-Model Benchmarking: Run side-by-side prompt comparisons across multiple models simultaneously.
• 💡 Reasoning / Thinking Blocks: Native rendering and collapsible sections for DeepSeek-R1 and other thinking models.
• 👥 AI Group Debate Mode: Multi-agent discussion loops (Optimist, Critic, Technologist) analyzing your prompts in real time.
• ⚡ AI IDE & Workspace: Full-screen code editor, persistent coding agent panel, version control (commits/reverts), and workspace export (.tar.gz).
• 🧠 RAG Knowledge Base: Upload .md or .txt files to chat with local documents.
• 💻 Built-in Sandbox Interpreter: Test and preview generated HTML/JS/CSS code locally.
• 🏪 Assistant Store: Built-in marketplace for free and premium system prompt presets.
• ⌨️ Full Vim Navigation Support: Toggle Vim keys anytime with F12.

---

🎹 Quick Keymap Cheat Sheet

Screen	Key Actions
Global	1..6 Switch screens · ? Help dialog · q Quit · F12 Toggle Vim mode
Chat	Enter Send · F2 New session · F7 Multi-model · F8 Debate · F9 Sandbox last code
IDE	F1 Files · F2 Versions · F4 New file · F5 Save · F9 Commit · PgUp/PgDn Scroll
Notes	F2 New note · F4 AI complement · F6 Export to RAG · F8 Preview
RAG	F2 Upload .md/.txt · Tab Switch focus · Enter Chat with docs

---

Licensed under AGPLv3. Full source code and documentation available in the repository.