Using Bridge
Bridge lives in the Bridge section of your project sidebar. Click it to open the conversational interface. This guide covers everything you need to know to use Bridge effectively.
Opening Bridge
- Click Bridge in the left sidebar (chat bubble icon).
- If this is your first time, a welcome message greets you.
- You're ready to start typing.
Bridge automatically loads the context for your current project — it already knows about your agents, workflows, and project settings.
Starting a Conversation
Type anything in the input field and press Enter (or click Send). Bridge understands natural language — you don't need special syntax.
Example opening messages:
"What agents do I have in this project?"
"Create an agent that searches the web and summarizes findings"
"Show me all failed executions from the last 24 hours"
"How do I set up a webhook trigger?"
"Run the content pipeline with the topic 'AI in manufacturing'"
Message Types
Bridge conversations contain several types of messages:
Your Message
Your typed text. Bridge reads it and responds.
Bridge Text Response
Conversational text from Bridge — explanations, lists, answers to questions. Formatted in markdown (headers, bullet points, code blocks).
Action Preview Card
When Bridge is about to make a change (create, update, delete), it shows a preview card before doing anything:
┌─────────────────────────────────────────────────────────────┐
│ ⚙️ Create Agent │
│ │
│ Name: researcher │
│ Model: claude-sonnet-4-6 │
│ API Key: My Claude Key │
│ Capabilities: web_search ✓, web_scraper ✓ │
│ System Prompt: "You are a research analyst..." │
│ │
│ [✓ Confirm] [✏️ Modify] [✗ Cancel] │
└─────────────────────────────────────────────────────────────┘
Click Confirm to proceed, Modify to adjust the proposed config before creating, or Cancel to abort.
Execution Status Card
When Bridge runs a workflow, it shows a live status card:
┌─────────────────────────────────────────────────────────────┐
│ ▶ Running: Content Pipeline │
│ Execution ID: exec_abc123 │
│ │
│ ● Trigger ✅ Complete (0.01s) │
│ ● researcher 🔄 Running... │
│ ● writer ⏳ Waiting │
│ ● Output ⏳ Waiting │
│ │
│ [View in canvas] │
└─────────────────────────────────────────────────────────────┘
This card updates in real time as nodes complete, in both Safe and Auto modes, and persists when you reload the conversation. The run uses the workflow's trigger default parameters unless you override them in your message (e.g. "run it with topic: AI in manufacturing").
Human-in-Loop Review Card
If a running workflow reaches a HiL node, Bridge shows the review card inline:
┌─────────────────────────────────────────────────────────────┐
│ 🔔 Approval Needed: "Review AI-generated blog post" │
│ │
│ The AI writer produced this post for your review: │
│ ─────────────────────────────────────────────────── │
│ ## Quantum Computing: A Beginner's Guide │
│ Quantum computing harnesses quantum mechanical... │
│ [... more content ...] │
│ ─────────────────────────────────────────────────── │
│ │
│ Comment (optional): ________________________ │
│ │
│ [✅ Approve] [❌ Reject] │
└─────────────────────────────────────────────────────────────┘
Conversation History
The left sidebar shows all your conversations, grouped by date:
▼ Today
Research Pipeline Setup
Agent Configuration
▼ Yesterday
Security Monitor
Content Pipeline v2
▼ Last Week
[5 more conversations]
Click any conversation to continue it. Bridge remembers the full context of each conversation.
Context Within a Conversation
Bridge maintains context across messages within a conversation:
You: What's the best model for code review tasks?
Bridge: For code review, I'd recommend claude-sonnet-4-6 or gpt-4o.
Claude tends to give more detailed explanations...
You: Ok, create an agent with that model
Bridge: [Creates agent with claude-sonnet-4-6 — because it inferred
"that model" from the previous message]
Bridge remembers what was discussed earlier in the conversation. After 30 messages, older messages are summarized and compressed to maintain context while staying within the model's context window.
Starting a New Conversation
Click + New Conversation at the top of the sidebar. This clears the context and starts fresh. Use new conversations when:
- Switching to a completely different topic
- You want Bridge to approach something without prior conversation context
- Previous conversation has exceeded 30 messages and context is getting compressed
Interrupting Bridge
If Bridge is generating a long response, click Stop (✗ button next to the Send button) to interrupt it. Bridge stops generating at the current point.
File Attachments
You can attach files to Bridge messages:
- Click the 📎 attachment icon next to the input field (or drag and drop).
- Supported formats:
.txt,.md,.json,.csv,.py,.js,.ts,.yaml - Bridge reads the file content as part of your message.
Use cases:
- "Here's a CSV of topics: [file]. Create a workflow that processes each one."
- "Review this Python script and create an agent that implements it."
- "Here's the JSON of a workflow I exported. Import it and adjust the agent names."
File size limit: 100KB per attachment.
Tips for Effective Bridge Use
Be Specific About Names
Bridge uses agent/workflow names to identify resources. The more specific you are, the better:
# Vague (might affect wrong agent)
"Update the researcher agent"
# Specific (clear which one)
"Update the gpt4o-researcher agent to use claude-sonnet-4-6 instead of gpt-4o"
Confirm Before Destructive Actions
Even in Auto Mode, Bridge always confirms deletes. Take a moment to read what it's about to delete before clicking Confirm.
Chain Actions in One Message
Bridge handles multi-step requests:
"Create an agent named analyzer with the security-analyst skill and web_search enabled,
then add it to the security-monitor workflow after the log-fetcher node"
Bridge will:
- Create the agent
- Open the workflow
- Add the node
- Connect it appropriately
Ask Follow-Up Questions
After Bridge does something, ask about it:
"Create a researcher agent with web search"
→ Bridge creates it
"Now test it with the query 'latest AI safety research'"
→ Bridge runs a test in the agent's test console
"The output was too long, add an instruction to keep responses under 200 words"
→ Bridge updates the system prompt
What Bridge Cannot Do
Bridge is powerful but has limits:
| Limitation | Workaround |
|---|---|
| Cannot access external URLs directly | Use an agent with web_search/web_scraper |
| Cannot read your LLM API key values | Keys are encrypted — view key names in Settings |
| Cannot modify account authentication settings | Use the TARX Settings page directly |
| Cannot manage billing or account-level settings | Use the relevant Settings page |
| Cannot access files on your local machine | Upload as attachments |
| Cannot browse the visual canvas | Describe what you want; Bridge adds nodes programmatically |
| Conversation limited to 30 messages | Start new conversation when compressed |