Langtrain
Langtrain Docs
DocsAPI ReferenceSDK Reference
AppChat
GitHubDiscord

Introduction

Langtrain is a comprehensive platform for building, fine-tuning, and deploying sovereign AI agents and models.

Visual IDE
Sovereign LoRA
Agentic CV
Python SDK

The Langtrain Ecosystem

Langtrain provides a complete suite of tools for modern AI development, allowing you to train custom models and orchestrate complex agents without compromising data privacy. Our ecosystem consists of four core pillars:
1. Langtrain Studio: A visual IDE for building AI agents and RLHF (Reinforcement Learning from Human Feedback) loops.2. Langtune Sandbox: A secure environment for sovereign LoRA fine-tuning.3. Langvision: Advanced agentic computer vision and screen control APIs.4. Langtrain SDK: A code-first framework in Python and TypeScript for programmatic access.

Langtrain Studio

Studio is our flagship Visual IDE designed for AI researchers and engineers.
  • •Agent Builder: Drag-and-drop interface for orchestrating multi-agent systems.
  • •RLHF Loops: Visually design reward models and human-in-the-loop feedback systems.
  • •Sovereign Execution: Run Claude Code, Codex, and Warp Agents securely within your VPC.

Langtune Sandbox

Langtune provides an isolated, sovereign environment for parameter-efficient fine-tuning (PEFT).
  • •LoRA & QLoRA: Train 70B+ parameter models on consumer hardware using 4-bit quantization.
  • •Data Privacy: Your training data never leaves your secure sandbox.
  • •Automated Evals: Built-in evaluation suites to prevent catastrophic forgetting during fine-tuning.

Langvision

Give your agents eyes and hands with Langvision.
  • •Agentic Computer Vision: State-of-the-art vision-language models capable of understanding complex UIs.
  • •Screen Control API: Allow agents to take autonomous actions on virtual desktops.
  • •Multi-modal Processing: Seamlessly combine text, image, and video inputs in your agent workflows.

Supported Models

We support the latest open-source models for fine-tuning and agent orchestration:
  • •Llama 3.3 by Meta (8B, 70B, 405B)
  • •DeepSeek V3 (70B, 236B)
  • •Qwen 2.5 by Alibaba
  • •Mistral/Mixtral by Mistral AI
All models are optimized for our proprietary training stack.

Next Steps

Ready to get started?
  • •Langtrain Studio - Build your first agent
  • •Langtune - Start sovereign fine-tuning
  • •Langvision - Explore computer vision capabilities
Next
Train in 10 Minutes