Skip to main content

Bridge Overview

Bridge is your AI command center — a conversational interface that lets you manage your entire TARX project through natural language. Instead of clicking through menus to create an agent, configure a workflow, or check execution status, you describe what you want in plain English and Bridge does it.

Bridge is powered by TARX's built-in AI model, provided at no cost to you. You don't need an LLM API key to use Bridge — it's included with every account.

What Bridge Knows​

Bridge has real-time access to your full project inventory:

  • Agents: Every agent you've configured — their names, models, capabilities, skills, RAGs, and MCPs
  • Workflows: Every workflow — node structure, trigger type, last execution status
  • Executions: Recent execution history, pending HiL approvals, failures
  • RAG Sources: All configured vector databases and their connection status
  • MCP Servers: All connected MCP servers and their tool counts
  • Credentials: All stored credentials (names only, not values)
  • Skills: Available skills (builtin + custom)
  • Members: Project members and their roles
  • Documentation: TARX docs indexed for accurate guidance

This gives Bridge the context to answer questions like "What agents do I have?" and "Show me yesterday's failed executions" without you needing to navigate anywhere.

What Bridge Can Do​

Bridge can read, create, update, and delete almost everything in your project:

Read Actions (instant, no confirmation)​

  • List agents, workflows, executions, RAGs, MCPs, skills
  • Show workflow structure and node config
  • Check execution status and results
  • Show pending Human-in-Loop approvals
  • Answer questions about TARX features

Write Actions (shows preview card before executing)​

  • Create agents, workflows, RAG sources, MCP server configs
  • Update agent config (model, system prompt, capabilities)
  • Create new nodes in workflows
  • Start workflow executions

Destructive Actions (always requires explicit confirmation)​

  • Delete agents
  • Delete workflows
  • Cancel running executions
  • Reject Human-in-Loop approvals

Two Modes​

Bridge has two confirmation modes for write actions:

Safe Mode (default)​

All write actions — create, update, execute — show a preview card before executing. You see exactly what Bridge is about to do (JSON preview of what will be created/changed) and click Confirm to proceed.

This mode is recommended for new users or when making important changes.

Auto Mode​

Non-destructive write actions (create, update) execute immediately without confirmation. Destructive actions (delete, reject) still require confirmation.

Use Auto Mode when you want Bridge to move faster — great for experienced users doing rapid iteration.

Toggle between modes with the switch in the Bridge header.

Bridge vs. Manual UI​

TaskManual UIBridge
Create an agentNavigate → Agents → New Agent → Fill form → Save"Create an agent named researcher with web_search and claude-sonnet-4-6"
Check yesterday's failuresNavigate → Executions → Filter by date + status"Show me yesterday's failed executions"
Run a workflowNavigate → Workflows → Open → Run → Enter input → Execute"Run the content pipeline with topic: AI in education"
Invite a team memberSettings → Projects → Members → Invite"Invite sarah@company.com as an editor to this project"
Update agent modelAgents → Open agent → Change model → Save"Change the researcher agent to use gpt-4o"

Bridge is faster for routine tasks. The manual UI is better for complex configurations (like building a workflow node-by-node on the canvas).

The Bridge Interface​

┌─────────────────────────────────────────────────────────────┐
│  Bridge                              [Safe Mode ●] [Auto ○] │
├─────────────────┬───────────────────────────────────────────┤
│  Conversations  │                                           │
│                 │  ┌─────────────────────────────────────┐  │
│  > Today        │  │ 🤖  Bridge                          │  │
│    Research Q   │  │                                     │  │
│    Config upd.  │  │ Hi! I'm Bridge. I have access to    │  │
│                 │  │ your full project. What would you   │  │
│  > Yesterday    │  │ like to do today?                   │  │
│    Sec monitor  │  └─────────────────────────────────────┘  │
│    Agent test   │                                           │
│                 │  ┌─────────────────────────────────────┐  │
│                 │  │ 👤 You                              │  │
│                 │  │                                     │  │
│                 │  │ Create an agent called summarizer   │  │
│                 │  │ using gpt-4o-mini that summarizes   │  │
│                 │  │ text in 3 bullet points             │  │
│                 │  └─────────────────────────────────────┘  │
│                 │                                           │
│                 │  [Preview Card — click Confirm]           │
│                 │                                           │
│                 │  [Type a message...]         [Send ↵]    │
└─────────────────┴───────────────────────────────────────────┘

Left panel: Conversation history, grouped by date. Click any past conversation to continue it. Bridge maintains context within a conversation.

Right panel: The active conversation. Messages from you and Bridge, plus action preview cards.

Input: Type your message and press Enter or click Send. Bridge supports multi-line input (Shift+Enter for newlines).

Bridge + RAG: TARX Knowledge​

Bridge is connected to a searchable knowledge base containing:

  • All TARX documentation (the docs you're reading right now)
  • Common workflow patterns and examples
  • Error messages and their solutions

This means Bridge can answer "How do I set up a webhook trigger?" or "What's the difference between web_search and web_scraper?" with accurate, up-to-date information — not hallucinated guesses.

Conversation Management​

Bridge conversations are:

  • Stored securely in TARX
  • Scoped to your project
  • Retained indefinitely (you can view old conversations)
  • Searchable by date in the sidebar
  • Limited to 30 messages per conversation (older messages are summarized to stay within context limits)

To start fresh: Click + New Conversation in the Bridge sidebar.

Next Steps​

Last updated on