What is TARX?
TARX is a visual AI agent orchestration platform — like n8n, but built from the ground up for the age of large language models. Instead of wiring API integrations together, you connect intelligent agents, logic, and triggers on a canvas and let them collaborate to accomplish complex tasks.
It is open-source and BYOK (bring your own key): you supply your own Anthropic, OpenAI, Google, or other LLM API keys. TARX never charges you per token, never stores your keys in plaintext, and never routes your data through a third-party billing layer.
Who It's For
TARX is a general-purpose platform for anyone building with AI — individual professionals, small teams, and large companies alike. If your work involves reasoning over text, automating decisions, enriching data, or coordinating several AI steps, TARX gives you a visual place to design and run it.
It is not tied to one role or industry. People use it for marketing and content, research and analysis, customer operations, engineering and DevOps, finance, security, education, and countless other domains. The common thread is simple: multiple AI steps that need to work together reliably.
The Core Idea
Traditional automation tools move data between services. TARX moves reasoning between agents.
Each node on the canvas can be a full LLM agent with its own model, system prompt, tools (capabilities), retrieval sources (RAG), and external tool servers (MCP). Nodes pass their outputs to downstream nodes via a simple expression system:
{{trigger.output}} → Agent 1 → {{agent_1.output}} → Agent 2 → Output
You build pipelines that think, branch, loop, wait for human approval, and produce structured results — all without writing orchestration code.
What You Can Build
These are examples, not limits — TARX adapts to whatever multi-step AI work you need.
Multi-Agent Pipelines
Connect specialized agents in sequence or parallel. A research agent feeds a writer agent, which feeds an editor agent, which delivers the final result.
Automated Monitoring & Reporting
Schedule a workflow to pull data on a cadence, classify events with an agent, branch on the important findings, and post a structured report or push it to another system via an HTTP node.
Human-in-the-Loop Systems
Build flows that pause for a person at critical points. A workflow can do the heavy lifting automatically, then wait for a reviewer to approve or reject before continuing.
Bulk Content & Data Processing
Use a Loop node to iterate over hundreds of items, run each through an agent, and collect the results into a single structured output.
Conversational Automation (Bridge)
Chat with Bridge to manage your entire project — create agents, build workflows, run executions — all in natural language. Bridge knows your project's full inventory and can take actions with a single confirmation.
Data Visualization (Canvas)
Turn workflow execution data into interactive charts. Describe the chart you want in plain English and Canvas generates it from your real run data.
How It Works
TARX runs as a cloud-hosted platform. Your browser connects to the TARX API, which orchestrates agent execution, streams live canvas updates, and routes LLM calls to your own providers.
| Layer | Technology | Notes |
|---|---|---|
| Frontend | React, Vite, TypeScript, React Flow | Visual canvas + Bridge chat |
| Backend | FastAPI (Python) | REST API, async executor, live streaming |
| Sign-in | Email + Google | Secure browser sessions handled by TARX |
| LLM routing | LiteLLM | Unified interface, your key injected per call (BYOK) |
Live execution updates stream to the canvas in real time, so you can watch each node start, run, and finish as a workflow executes.
BYOK: Bring Your Own Key
TARX's model is not to charge you per token. You bring your own API keys and TARX injects them into LLM calls on your behalf. Supported providers include:
- Anthropic — Claude models
- OpenAI — GPT models
- Google AI — Gemini models
- Mistral
- HuggingFace — open-source hosted models
- Azure AI Foundry — your own Azure deployments
- OpenAI-compatible — any OpenAI-spec endpoint (Groq, Together, OpenRouter, Ollama, vLLM, and more)
The exact models available are listed live from your key when you configure an agent — you don't need to memorize model strings. Your keys are encrypted at rest, decrypted in-memory only at execution time, and never logged or returned in plaintext. See BYOK: API Keys.
Platform Concepts
Projects
Everything in TARX lives inside a project. A project has a URL slug, a team of members with roles (admin / editor / reader), and its own agents, workflows, credentials, RAG sources, and MCP servers. Your URL structure is always /{username}/{projectSlug}/{page}.
Agents
An agent is a reusable LLM configuration — model, system prompt, temperature, capabilities, skills, RAG sources, and MCP servers. Agents live in your project and can be reused across multiple workflows.
Workflows
A workflow is a directed graph of nodes on the visual canvas. When it executes, nodes run in order, with each node's output available to downstream nodes via expressions like {{agent_1.output}}.
Bridge
Bridge is your AI command center — a chat interface powered by TARX's built-in AI model (no user key needed). Bridge has full knowledge of your project and can take actions: create agents, build workflows, run executions, and more. You can also wire your own agents into Bridge as delegates.
Canvas
Canvas is TARX's native visualization layer. Link a canvas to your workflows, attach a canvas agent (using your own key), and generate interactive charts from execution data in plain English.
Navigation
| Section | What you find there |
|---|---|
| Getting Started | Account setup, first project, adding keys, first workflow |
| Agents | Creating and configuring LLM agents |
| Workflows | Canvas, node types, triggers, expressions, execution |
| Canvas | Data visualization from workflow runs |
| Bridge | AI command center, actions, multi-agent orchestration |
| RAG Sources | Vector database connections, embedding strategy |
| MCP Servers | Model Context Protocol server connections |
| Skills | Prompt presets for agents |
| Settings | Account, projects, keys, credentials |
| User Journeys | End-to-end walkthroughs of common use cases |
| Reference | Expression reference, node specs, API overview, limits |
If you just want to try TARX, follow the Quick Start guide. You'll have a working workflow in under 10 minutes.
You need at least one LLM API key to use agent nodes. Bridge works without a user key — it uses TARX's built-in AI model.