Building Workflows with Bridge
One of Bridge's most powerful features is its ability to translate natural language descriptions into complete, connected workflow graphs — ready to run on the canvas.
Instead of manually dragging nodes, connecting edges, and configuring each field, you describe what you want and Bridge plans the workflow, shows you a preview, and creates it with a single confirmation.
How It Works
Step-by-Step: Creating a Workflow
Step 1: Describe What You Want
Be as specific or as high-level as you like. Bridge asks clarifying questions if needed.
High-level:
"Create a workflow that monitors our Slack channel for customer complaints,
analyzes sentiment, and creates a Jira ticket if it's negative"
Detailed:
"Create a workflow with:
- A webhook trigger that receives Slack events
- An agent named sentiment-analyzer that classifies the message as positive/negative/neutral
- A branch node that checks if sentiment is negative
- If negative: an agent that creates a Jira ticket with the complaint details
- An output node"
Both work. The detailed version gives Bridge more precision, but the high-level version is faster if you're comfortable editing on the canvas afterward.
Step 2: Bridge Plans the Workflow
Bridge thinks through the node structure and shows a plan:
Bridge: Here's the workflow I'll create:
📋 Workflow Plan: Slack Sentiment Monitor
Nodes:
1. Webhook Trigger
- Receives Slack Event Subscription payload
- Extracts: message text from $.event.text
2. Agent: sentiment-analyzer
- Input: {{trigger.output}}
- System prompt: Classify sentiment as "positive", "negative", or "neutral"
- Output format: Single word
3. Branch: Is it negative?
- Condition: output.strip() == "negative"
4. [True] Agent: jira-creator
- Input: Create a Jira ticket for this complaint: {{trigger.output}}
- Uses the Jira MCP server to create the ticket
5. Output: {{jira-creator.output}} / "No action taken"
Agents needed:
- sentiment-analyzer (new) — gpt-4o-mini, classification task
- jira-creator (new) — attach the Jira MCP server
Shall I create this workflow and the two new agents?
[✓ Create Workflow + Agents] [✏️ Modify Plan] [✗ Cancel]
Step 3: Confirm or Modify
Confirm: Bridge creates everything — the workflow structure and any new agents.
Modify Plan: An inline editor appears showing the planned structure. Adjust nodes, change agents, modify connections, then confirm.
Step 4: Open in Canvas
After creation:
Bridge: ✅ Created!
Workflow: "Slack Sentiment Monitor" is ready
New agents created: sentiment-analyzer, jira-creator
Note: The jira-creator agent needs your Jira API endpoint and
credentials — configure these in the agent's system prompt.
[Open in Canvas →] [Configure jira-creator Agent →]
Click Open in Canvas to see the workflow visually. You can adjust positions, review configurations, and make final tweaks.
Example Workflow Requests
Simple Pipeline
"Build a workflow that takes a URL as input, scrapes the page, summarizes it,
and outputs the summary"
Bridge creates:
[Manual Trigger] → [Agent: scraper (web_scraper)] → [Agent: summarizer] → [Output]
Scheduled Report
"Create a daily workflow that runs at 9am EST, fetches our API health endpoint,
and emails me if the status is not 'ok'"
Bridge creates:
[Schedule Trigger: 0 9 * * * EST] → [http_request node: GET health endpoint]
→ [Branch: status != ok?] → [True: http_request node: send alert] → [Output]
Bulk Processing
"Create a workflow that accepts a JSON array of topics, writes a 200-word blog
intro for each one, and outputs all the intros"
Bridge creates:
[Manual Trigger] → [Loop: items={{trigger.output}}]
→ [Agent: writer (input: {{loop.current_item}})] → [Output: {{loop.output}}]
Complex Multi-Agent with HiL
"I need a content production pipeline:
- Webhook receives a content brief
- Researcher agent searches the web for relevant info
- Writer agent creates a draft
- Editor agent refines it
- Human reviews the final draft
- If approved, post to our CMS API
- If rejected, send back for revision"
Bridge creates a 7-node workflow with all the connections and a revision loop.
Iterating with Bridge
After creating a workflow, continue the conversation to refine it:
You: "Change the summarizer agent to use claude-haiku-4-5 instead"
Bridge: ✅ Updated — researcher agent now uses claude-haiku-4-5
You: "Add a note node explaining what the branch condition does"
Bridge: ✅ Added a note node: "Branch routes long content (>500 chars) to
summarizer, short content to formatter"
You: "Set the webhook trigger's payload path to extract just the body.message field"
Bridge: ✅ Updated webhook trigger — payload_path set to $.body.message
What Bridge Creates vs. What You Configure
Bridge creates the structure — nodes, connections, basic config. Some things still need manual configuration:
| Bridge Creates | You Configure |
|---|---|
| All node types and connections | Agent's API key selection |
| Basic system prompts | Webhook secret (generated, but you copy it to your service) |
| Capability toggles (based on task) | Cron timezone and exact timing |
| Agent names and models | MCP server connections (project-specific) |
| Input expressions | Jira/Slack/external API credentials |
Bridge will tell you what still needs configuration after creating the workflow.
Workflow Preview Mode
Before creating, you can ask Bridge to show a visual preview:
"Before creating, show me what the security monitor workflow would look like"
Bridge: Here's the planned structure:
[Schedule: 0 */4 * * *] ──→ [log-fetcher] ──→ [security-analyzer]
──→ [Branch: severity == critical?]
├── True ──→ [escalator] ──→ [Output: Alert sent]
└── False ──→ [logger] ──→ [Output: Logged]
Total: 6 nodes, 5 connections
Agents to create: log-fetcher, security-analyzer, escalator, logger
Does this look right? I can adjust before creating.
You can request changes to the plan before any creation happens.
Tips for Effective Workflow Creation
Name Your Agents
If you reference agents by name that already exist in your project, Bridge uses them:
"Create a workflow using my existing researcher agent, then a writer agent, then output"
If they don't exist, Bridge creates them with reasonable defaults.
Specify Trigger Details
Be specific about how you want workflows to start:
# Vague
"Create a scheduled workflow..."
# Specific
"Create a workflow that runs every weekday at 7am Eastern..."
Mention Required Tools
If you know the workflow needs to call an external API or service, say so — Bridge will use an http_request node or the right MCP server:
"Add an http_request node that calls our internal API and passes the result to the analyzer"
Include Output Format Requirements
If downstream systems need specific formats:
"The final output should be a JSON object with fields: title, summary, tags, severity"