Complete Platform Capabilities
Bring Your Own Models
Connect AI providers or any OpenAI-compatible endpoint using your own API keys. Configure
model ID, base URL, temperature, max tokens, system prompt, and more. AiBenchX includes
presets for OpenAI, Anthropic, Google Gemini, Mistral, Groq, OpenRouter, DeepSeek,
Kimi/Moonshot, Z.AI/GLM, Qwen/DashScope, Perplexity, xAI, Together AI, Fireworks,
DeepInfra, Cerebras, Hugging Face, NVIDIA NIM, MiniMax, Pixazo (image generation),
Ollama (local), LM Studio (local), and any custom OpenAI-compatible endpoint. Each
connected provider appears in the composer alongside built-in models — LFM, Kat, Llama,
and Laguna M.1 — so switching models is part of the natural workflow.
Thinking Mode Controls
For models that support reasoning — such as OpenAI o-series, Claude, Gemini 2.5+,
DeepSeek R1, and Qwen3 — AiBenchX offers three thinking modes: Think Auto, Think On, and
Think Off. Choose how much reasoning depth you need for each task without learning every
provider's naming conventions.
Build Mode
A full code workspace with file tree, syntax-highlighted editor, live HTML preview,
per-turn undo history, and one-click ZIP export. When the conversation should produce
working files — not just advice — toggle Build Mode and let the AI generate code directly
into an editable project. It works especially well for landing pages, front-end
prototypes, small tools, interactive demos, and HTML experiments.
Expert Channels
Nine specialized AI workflows that bring focused expertise to writing, research, strategy,
learning, and professional communication. Instead of starting every conversation with a
blank assistant, Experts give you a purpose-built AI workflow with a defined specialty,
step-by-step methodology, clear quality standards, and stated boundaries. The selected
model still decides the provider endpoint — the expert decides the working style.
- AI Humaniser — rewrites text with the natural rhythm, varied cadence, and genuine craft of an experienced human writer — every sentence composed, not corrected.
- Expert Content Writer — plans and drafts structured, evidence-backed content with a clear angle, strong transitions, and a voice tailored to your target audience.
- Research Analyst — breaks questions into claims, evidence, and uncertainty — separating what is established from what still needs verification, with honest confidence levels.
- Study Coach — teaches topics step by step with guided explanation, active recall, and practice problems — adapting to your level and pace as you progress.
- Business Strategist — frames decisions as structured trade-offs with explicit assumptions, second-order effects, risk assessment, and a concrete recommended next move.
- Resume and Career Coach — sharpens resumes, interview stories, and professional positioning using your real experience — truthful, specific, and recruiter-ready.
- SEO Writer — creates content that answers what your audience is actually searching for — useful depth first, search visibility second, never keyword-padded.
- Email and Reply Writer — drafts clear, concise emails and replies in exactly the right tone — with one clear ask and nothing left to misinterpret.
- Code Architect — engineers production-grade code with complete error handling, edge case coverage, security awareness, performance consideration, and verification before delivery.
Rich Response Rendering
AI responses render with full Markdown support, GitHub-flavored tables and task lists,
inline and block LaTeX/KaTeX math notation, syntax-highlighted code blocks (30+
languages), Mermaid diagrams with SVG export, HTML live previews, copy controls, and
auto-balancing math delimiters during streaming. This makes AiBenchX especially valuable
for engineering documentation, study explanations, mathematical formulas, technical
drafts, diagrams, and generated web code.
File Attachments
Attach images and documents to add visual or file-based context. Vision-capable models can
analyze image uploads directly. Document attachments contribute readable text content as
context.
Image Generation
Generate images from text prompts using any configured image-capable provider. View,
download, and keep generated images in your conversation. Image generation uses providers
configured specifically for image output — standard text models produce text, image models
produce images. Your generated images are stored durably so they persist across sessions.
Voice Playback
Listen to any AI response as natural speech with true streaming that begins playing before
the response finishes. Code blocks, formatting syntax, and other non-narration content are
automatically stripped for clean, natural speech. Ideal for reviewing long explanations,
listening to drafted articles to catch tone issues, study sessions, and accessibility.
PDF Export
Export any AI response as a styled PDF document — ready to share, archive, or print.
Useful for research summaries, study notes, lesson explanations, technical answers, and
client-ready drafts.
User Adaptation
AiBenchX adapts to your preferences, tone, and working context within each conversation —
so follow-up responses get sharper over time. Adaptation is scoped to the current thread,
so each conversation maintains its own context independently. Disable or re-enable
anytime from Settings, or delete the adaptation for the current thread when you want a
fresh start.
Threaded Conversations
Keep every project, topic, and experiment in its own thread with rename, delete, and
instant access to your full conversation history. Expert-started threads display a visual
indicator so you can identify which expert workflow was used. Threads receive automatic
titles based on conversation content.
Character Companion
An optional animated companion that reacts to your workspace and offers curated wisdom
quotes from the Bhagavad Gita in Sanskrit, Hindi, and English. Choose Auto mode (quotes
match the workspace moment) or Manual mode (select mood categories). Customizable through
Settings.
Theme Support
Dark, light, or system-matched appearance. Your preference persists across sessions and
applies across the entire workspace.