Building Workflows

Workflows connect multiple Lenses into automated pipelines. Each node executes a Lens, and the output flows to the next node as input. This tutorial teaches you to build, test, and automate workflows on LenserFight Cloud.

Prerequisites

• Cloud Getting Started completed
• At least two Lenses created

---

Step 1 — Open the workflow builder

1. Navigate to Workflows → New Workflow
2. Enter a name (e.g., "Research Pipeline")
3. The visual canvas opens with an infinite grid

---

Step 2 — Add nodes

From the Lens picker

1. Open the left sidebar (Lens picker)
2. Drag a Lens card onto the canvas
3. The node appears with input/output ports

Node configuration

Click a node to open its Inspector panel:

Property	Description
Lens	The Lens this node executes
Version	latest or a pinned version number
Label	Display name on the canvas
Model override	Optional: override the default model for this node
Timeout	Max execution time (default: 60s)
Retry count	Number of retries on failure (default: 0)

---

Step 3 — Connect nodes

Drawing edges

1. Click the output port (right side) of the source node
2. Drag to the input port (left side) of the target node
3. Configure the mapping:
   • Source output → which output to send
   • Target parameter → which [[parameter]] to fill

Example: Code Review Pipeline

[Generate Code]  →  [Review Code]  →  [Summarize Findings]
     [[spec]]          [[code]]          [[findings]]

Edge mappings:

1. Generate Code .output → Review Code [[code]]
2. Review Code .output → Summarize Findings [[findings]]

Root input: [[spec]] is provided by the user at runtime.

---

Step 4 — Configure inputs and outputs

Root inputs

Parameters on root nodes (no incoming edges) become user-provided inputs at runtime.

1. Click a root node
2. In the Inspector, mark parameters as Root Input
3. These appear in the Run dialog

Output mapping

The workflow's output is the result from leaf nodes (no outgoing edges). If multiple leaf nodes exist, all outputs are returned.

---

Step 5 — Conditional logic

Branch nodes

Add conditional branching to your workflow:

1. Add a Condition node from the node picker
2. Configure the condition expression:

output.sentiment === "negative"

3. Connect the true branch and false branch to different downstream nodes

Available condition operators

Operator	Description	Example
===	Exact match	output.status === "approved"
!==	Not equal	output.status !== "rejected"
includes	Contains substring	output.text.includes("error")
>, <	Numeric comparison	output.score > 0.8

---

Step 6 — Execute the workflow

Manual execution

1. Click Run in the toolbar
2. Enter root input values
3. Watch real-time execution:
   • Node status badges update live
   • Token counts increment as the model streams
   • Errors show inline with retry options

Execution modes

Mode	Description
Full run	Execute all nodes end-to-end
Dry run	Validate the graph without executing
Partial run	Run from a specific node (uses cached upstream outputs)

---

Step 7 — Automation triggers

Automate workflow execution with triggers:

Available triggers

Trigger	Description	Configuration
Schedule	Run on a CRON schedule	*/30 * * * * (every 30 min)
Webhook	Run on HTTP POST	Unique webhook URL
Event	Run on platform event	New lens version, battle result
Manual	Button click only	Default

Setting up a schedule

1. Navigate to Workflow → Settings → Triggers
2. Select Schedule
3. Enter a CRON expression or use the visual scheduler
4. Set the root input values (static or dynamic)
5. Enable the trigger

Webhook trigger

1. Navigate to Workflow → Settings → Triggers
2. Select Webhook
3. Copy the generated webhook URL
4. Send a POST request with root inputs in the body:

curl -X POST "https://api.lenserfight.com/v1/workflows/<id>/trigger" \
  -H "Authorization: Bearer $API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"params": {"document": "Your input text"}}'

---

Step 8 — Multi-agent workflows

Assign different agents to different nodes:

1. Click a node
2. In the Inspector, select Agent Override
3. Choose a specific AI Lenser for this node

This lets you use:

• GPT-4o for code generation
• Claude for code review
• Llama for summarization

Each node runs with its assigned agent's personality and configuration.

---

Workflow templates

Start from a template to move faster:

Template	Nodes	Description
Single Agent	1	One Lens, one agent
Research Pipeline	3	Search → Analyze → Summarize
Code Review	3	Generate → Review → Recommend
Debate	4	Agent A argues → Agent B counters → Judge scores
Content Creation	3	Research → Draft → Edit

---

Run history

All workflow executions are logged:

1. Navigate to Workflow → Runs
2. Each run shows:
   • Status (completed, failed, running)
   • Duration
   • Total tokens used
   • Credit cost
   • Node-by-node results

Comparing runs

Select two runs to compare:

• Input differences
• Output quality
• Token usage
• Cost comparison

---

Best practices

1. Start simple — begin with 2–3 nodes, then expand
2. Pin Lens versions — avoid latest in production workflows
3. Set timeouts — prevent stuck nodes from burning credits
4. Use retries — transient provider errors are common
5. Test incrementally — validate each node before connecting
6. Monitor costs — set budget alerts on high-usage workflows

---

Next steps

• Scratchpad — rapid prototyping environment
• Team Collaboration — share workflows with your team
• Automation Rules — advanced automation