🔄Workflow Management

Complete guide to creating, executing, and monitoring multi-agent workflows

Orchestrate complex automation with intelligent multi-agent workflows

---

📖 Table of Contents

1. Overview

2. Quick Start

3. Active Tab

4. Templates Tab

5. History Tab

6. Monitoring Tab

7. Live Tab (Execution Theater)

8. Creating Workflows

9. Executing Workflows

10. Common Tasks

11. Advanced Features

12. Troubleshooting

---

Overview

What is Workflow Management?

Workflows are coordinated sequences of agents working together to accomplish complex tasks. Think of workflows as projects where multiple AI agents collaborate like a team.

Access: Navigate to Workflows from the main sidebar

Workflow Management Overview

What Can You Do Here?

• ✅ Create workflows with multiple coordinated agents

• ✅ Monitor execution in real-time with live progress

• ✅ Use templates for common automation patterns

• ✅ Track history of all workflow executions

• ✅ Optimize performance with detailed monitoring

• ✅ Watch live with the Execution Theater

Page Layout

The Workflows page has 5 main tabs:

1. ▶️ Active - Currently running workflows

2. 📋 Templates - Reusable workflow patterns

3. 🕐 History - Past workflow executions

4. 📊 Monitoring - Real-time performance metrics

5. 🎬 Live - Execution Theater (immersive view)

---

Quick Start

Running Your First Workflow (5 Minutes)

Goal: Execute a pre-built workflow template

Steps:

1. Go to Templates tab

2. Find "Code Review Workflow" template

3. Click "Use Template"

4. Enter workflow details:

   • Name: "Review PR #123"

   • Context: Add your repository info

5. Click "Create & Execute"

6. Switch to Live tab to watch execution

⏱️ Time: 5 minutes 🎯 Result: Workflow completes with results

See detailed walkthrough →

---

Active Tab

Overview

Shows all currently running or pending workflows.

💡 Tooltip: "Monitor active workflows with real-time progress updates"

Active Tab

Statistics Cards

🔄 Active Workflows 💡 Tooltip: "Workflows currently executing. Includes running and queued workflows."

• Count of active + queued workflows

• Changes from previous period

• Example: "3 workflows"

✅ Completed Today 💡 Tooltip: "Successfully completed workflows in the last 24 hours"

• Today's completion count

• Success rate percentage

• Example: "12 completed"

⏱️ Avg Duration 💡 Tooltip: "Average time from start to completion. Helps estimate new workflows."

• Average execution time

• Trend indicator

• Example: "5m 23s"

🤖 Agent Utilization 💡 Tooltip: "Percentage of agents currently working on workflows. Higher = busier system."

• Active agents / Total agents

• Utilization percentage

• Example: "67%"

Active Tab Stats

Workflow Cards

Each active workflow shows:

Header:

• Workflow Name

• Status Badge:

  • 🔵 Running (blue)

  • 🟡 Queued (yellow)

  • 🟢 Completing (green)

  • 🔴 Error (red)

• Elapsed Time: Time since start

Progress Indicator:

💡 Tooltip: "Visual progress through workflow stages. Each segment represents a task or agent."

• Progress bar with percentage

• Current stage highlighted

• Estimated time remaining

Agents Involved:

• Icons/avatars of participating agents

• Agent status indicators

• Current agent highlighted

Workflow Card

Live Metrics:

• Subtasks: Completed / Total

• Quality: Current quality score

• Tokens: Tokens used so far

• Cost: Estimated cost

Action Buttons:

• 👁️ Watch: Open in Execution Theater

• ⏸️ Pause: Temporarily halt execution

• ⏹️ Stop: Cancel workflow

• 📄 View Details: Open detailed modal

Quick Actions

Create Workflow: Opens Create Workflow Modal

Refresh: Reload active workflows (auto-refreshes every 5s)

Filter: Filter by status, agent, or tags

---

Templates Tab

Overview

Browse and use pre-built workflow templates for common automation patterns.

💡 Tooltip: "Templates are proven workflow patterns. One-click creation with your custom settings."

Templates Tab

Template Categories

🔒 Security & Compliance:

• Security Audit Workflow

• Compliance Check Workflow

• Vulnerability Scan Workflow

💻 Development:

• Code Review Workflow

• PR Analysis Workflow

• Documentation Generation Workflow

📊 Data & Analytics:

• Data Analysis Workflow

• Report Generation Workflow

• Insight Discovery Workflow

🚀 DevOps:

• Deployment Workflow

• Infrastructure Audit Workflow

• Performance Testing Workflow

Template Categories

Template Cards

Each template shows:

Template Information:

• Name: Template identifier

• Category Badge: Color-coded category

• Description: What the workflow does

• Usage Count: How many times used

• Success Rate: Historical success %

• Avg Duration: Typical execution time

• Avg Cost: Typical cost

Template Card

Agent Pattern:

Shows the sequence of agents:

CodeArchitect → SecurityExpert → DocumentGenerator

💡 Tooltip: "Agent execution order. Arrows indicate sequential execution, parallel shown side-by-side."

Performance Stats:

• ⏱️ 4m 5s avg execution time

• 💰 $0.12 avg cost

• 🎯 23 uses historical

• ⭐ 95.7% success rate

Actions:

• Use Template: Create workflow from this template

• Preview: View template details

• Customize: Modify before creation

Using a Template

Steps:

1. Click "Use Template" on desired template

2. Workflow Name: Enter custom name

3. Description (optional): Describe this instance

4. Context: Provide task-specific information

   {
     "repository": "github.com/myorg/myapp",
     "branch": "main",
     "pr_number": 123
   }

5. Customize Agents (optional):

   • Add extra agents

   • Remove agents

   • Change agent order

6. Click "Create Workflow"

⏱️ Time: 2 minutes 🎯 Result: Workflow created and ready to execute

---

History Tab

Overview

View all past workflow executions with detailed results.

💡 Tooltip: "Complete execution history. Review past workflows, analyze patterns, learn from failures."

History Tab

History Filters

Time Range 💡 Tooltip: "Filter by execution date. Narrow down to find specific workflows."

• Last 24 hours

• Last 7 days

• Last 30 days

• Last 90 days

• Custom range

Status Filter 💡 Tooltip: "Filter by execution outcome"

• All: Show everything

• ✅ Completed: Successful executions

• ❌ Failed: Failed executions

• ⏸️ Paused: Manually paused

• ⏹️ Cancelled: Manually stopped

Agent Filter 💡 Tooltip: "Show only workflows that used specific agents"

• Select agent from dropdown

• Shows all workflows involving that agent

Search:

• Search by workflow name

• Search by description

• Search by tags

History Filters

Execution Records

Each history entry shows:

Execution Summary Card:

• Workflow Name: Original workflow name

• Status Badge: Final status with color

• Execution ID: Unique identifier

• Started: Start timestamp

• Duration: Total execution time

• Quality Score: Overall quality (0.0-1.0)

• Cost: Total cost for execution

History Record

Agents Involved:

• Icons of all participating agents

• Agent status during execution

• Time each agent was active

Results Summary:

• Subtasks completed

• Success rate

• Issues encountered

• Output artifacts

Actions:

• 📄 View Results: See detailed execution results

• 🔄 Re-run: Execute same workflow again

• 📊 Compare: Compare with other executions

• 💾 Export: Download execution data

Execution Details

Click on any execution to see full details:

Detailed Results Modal:

💡 Tooltip: "Complete breakdown of workflow execution with all agent outputs"

Shows:

1. Timeline: Visual timeline of execution

2. Agent Outputs: Results from each agent

3. Quality Scores: Breakdown by subtask

4. Resource Usage: Tokens and costs

5. Logs: Complete execution logs

6. Artifacts: Generated files/outputs

Execution Details Modal

---

Monitoring Tab

Overview

Real-time monitoring of workflow internals - memory, RAG, tools.

💡 Tooltip: "Advanced monitoring of workflow execution internals. See how context engineering and tools are being used."

Monitoring Tab

Sub-Tabs

The Monitoring tab has 3 sub-tabs:

1. Memory

What it shows:

• Memory items created during execution

• Working memory state

• Memory consolidation events

• Cross-agent memory sharing

Memory Activity Feed 💡 Tooltip: "Real-time stream of memory operations during workflow execution"

Each entry shows:

• Timestamp: When memory was stored/retrieved

• Agent: Which agent accessed memory

• Operation: Store, Retrieve, Update, Delete

• Memory Type: Experience, Knowledge, Skill, Tool Usage

• Content Preview: First 100 characters

• Importance: Score 0.0-1.0

Monitoring Memory Tab

Memory Statistics:

• Total items stored

• Memory by tier (working/short-term/long-term)

• Memory by agent

• Consolidation events

2. RAG (Retrieval-Augmented Generation)

What it shows:

• Document retrievals during execution

• RAG query performance

• Context chunks used

• Retrieval quality scores

RAG Activity Feed 💡 Tooltip: "Track when agents search knowledge base for context"

Each RAG query shows:

• Agent: Who made the query

• Query: Search text

• Results: Number of chunks retrieved

• Latency: Time taken

• Relevance: Average similarity score

• Tokens: Token count of retrieved context

Monitoring RAG Tab

RAG Performance:

• Total queries made

• Average latency

• Cache hit rate

• Token efficiency

Retrieved Chunks Preview:

• Source documents used

• Chunk text previews

• Similarity scores

• How chunks were used

3. Tools

What it shows:

• External tools called during execution

• Tool call success/failure

• Tool response times

• Integration errors

Tool Activity Feed 💡 Tooltip: "Monitor external tool usage. Identify slow tools or failing integrations."

Each tool call shows:

• Agent: Who called the tool

• Tool Name: Which integration

• Method: What action (e.g., "createPullRequest")

• Status: ✅ Success, ❌ Failed, ⏳ Pending

• Duration: Time taken

• Result Preview: Output summary

Monitoring Tools Tab

Tool Statistics:

• Total tool calls

• Success rate by tool

• Average latency by tool

• Failed tool calls (with errors)

Tool Performance Alerts:

• 🔴 Slow tools (>5s response)

• 🔴 Failing tools (>10% failure rate)

• 🟡 Rate-limited tools

• 🟢 Healthy tools

---

Live Tab (Execution Theater)

Overview

The Execution Theater is an immersive, real-time view of workflow execution.

💡 Tooltip: "Watch workflows execute in real-time with cinematic visualization. Perfect for demos and debugging."

Execution Theater

Theater Features

Full-Screen Mode:

• Click full-screen icon (top right)

• Immersive execution view

• Press Esc to exit

Agent Visualization:

Shows each agent as a card/avatar with:

• Agent name and icon

• Current status (idle/thinking/executing/complete)

• Progress indicator

• Current task description

• Thinking animation while processing

Theater Agent Cards

Execution Flow:

Visual representation of workflow stages:

1. Task Decomposition: LLM breaks down the goal

2. Agent Selection: System chooses optimal agents

3. Context Engineering: Building optimal prompts

4. Execution: Agents performing tasks

5. Aggregation: Combining results

6. Learning: Storing patterns

Each stage animates as it completes.

Live Logs Panel:

💡 Tooltip: "Real-time log stream. See exactly what each agent is doing."

• Scrolling log feed

• Color-coded by agent

• Filterable by log level

• Copy/export logs

Theater Live Logs

Performance Metrics (bottom bar):

• ⏱️ Elapsed time

• 🎯 Subtasks completed

• 🪙 Tokens used

• 💰 Estimated cost

• 📊 Quality score (updates as tasks complete)

Sub-Tabs in Theater

Overview (default):

• Full execution visualization

• All agents visible

• Combined progress

Memory:

• Memory operations in real-time

• Memory graph visualization

• Inter-agent memory sharing

Communication:

• Agent-to-agent messages

• Communication flow diagram

• Message history

Learning:

• Patterns being learned

• Knowledge being stored

• Performance improvements

Theater Sub-Tabs

---

Creating Workflows

Create Workflow Modal

Click "Create Workflow" button to open 2-step wizard.

Create Workflow Button

Step 1: Details

Workflow Name (required) 💡 Tooltip: "Descriptive name for this workflow. Include the goal or project identifier."

• Clear, specific names

• Include version if iterating

• Example: "Security Audit - Q4 2024"

Description (optional) 💡 Tooltip: "Explain what this workflow does and when to use it"

• What problem does it solve?

• What are the inputs/outputs?

• Any special notes

Goal (required) 💡 Tooltip: "The high-level objective. The orchestrator uses this to decompose tasks."

• Clear, specific goal statement

• Good: "Perform comprehensive security audit of authentication system checking for SQL injection, XSS, and authentication bypass vulnerabilities"

• Bad: "Review code"

Create Workflow Step 1

Tags (optional) 💡 Tooltip: "Organize workflows with tags. Helps filtering and searching."

• Add relevant tags

• Examples: "security", "production", "weekly"

• Suggestions based on goal

Category (optional):

• Security

• Development

• Data Analysis

• Infrastructure

• Custom

Step 2: Configuration & Policy

Agent Selection:

💡 Tooltip: "Choose which agents participate. Leave empty for automatic selection based on task requirements."

Manual Selection:

• Select specific agents from dropdown

• Define execution order

• Set agent roles (lead, support)

Automatic Selection (recommended): 💡 Tooltip: "Let the orchestrator choose optimal agents based on skills and availability"

• ✅ Enable "Auto-Select Agents"

• System analyzes goal

• Picks best agents automatically

• Uses LLM reasoning for selection

Create Workflow Step 2

Execution Policy:

💡 Tooltip: "Rules governing workflow execution behavior"

Execution Mode:

• Parallel: Run independent tasks simultaneously (faster)

• Sequential: One task at a time (safer)

• Optimized: Mix of parallel/sequential (recommended)

Error Handling:

• Stop on Error: Halt if any task fails

• Continue: Skip failed tasks, continue with rest

• Retry: Auto-retry failed tasks (1-5 attempts)

Timeout:

• Per Task: Max time for individual tasks (default: 5 min)

• Whole Workflow: Max total time (default: 30 min)

Resource Limits:

• Token Budget: Max tokens for entire workflow

• Cost Budget: Max cost in USD

• Agent Limit: Max agents to use

Context Configuration:

💡 Tooltip: "Add context that all agents can access during execution"

{
  "repository": "https://github.com/myorg/myapp",
  "codegraph_project": "myapp",
  "compliance_standards": ["SOC2", "GDPR"],
  "custom_data": {...}
}

Review & Create

Final Review:

• Workflow name and goal

• Selected/auto-selected agents

• Execution policy summary

• Estimated duration

• Estimated cost

Buttons:

• ← Back: Modify settings

• Create: Create workflow (don't execute)

• Create & Execute: Create and run immediately

---

Executing Workflows

Manual Execution

From History or Templates:

1. Find the workflow

2. Click "Run" or "Re-run" button

3. Confirm execution (optional input data)

4. Click "Execute Workflow"

Run Workflow

Execution Process

Phase 1: Initialization (2-5 seconds)

💡 Tooltip: "Setting up execution environment and loading agents"

• Load workflow definition

• Initialize agents

• Prepare execution context

• Queue workflow

Phase 2: Task Decomposition (5-15 seconds)

💡 Tooltip: "LLM analyzes the goal and breaks it into subtasks"

• Goal analyzed by orchestrator

• Subtasks identified

• Dependencies mapped

• Priorities assigned

Phase 3: Agent Selection (3-10 seconds)

💡 Tooltip: "Optimal agents chosen using 4D matching (skills, availability, history, tools)"

• Agent capabilities evaluated

• Best matches selected

• Assignments confirmed

• Agents notified

Phase 4: Context Engineering (5-20 seconds)

💡 Tooltip: "Optimizing context for each agent using RAG and memory retrieval"

• Relevant documents retrieved

• Agent memories loaded

• Context optimized for token budget

• Prompts generated

Phase 5: Execution (variable, 30s-10min)

💡 Tooltip: "Agents executing their assigned tasks. Progress updates in real-time."

• Agents work in parallel/sequential

• Tools called as needed

• Results shared between agents

• Progress updates sent

Phase 6: Aggregation (5-10 seconds)

💡 Tooltip: "Combining agent results into coherent final output"

• Results collected

• Quality scores calculated

• Conflicts resolved

• Final output generated

Phase 7: Learning (2-5 seconds)

💡 Tooltip: "Storing patterns and updating agent knowledge"

• Patterns extracted

• Successful strategies saved

• Agent memories updated

• Performance metrics recorded

Execution Phases

Monitoring Execution

Real-time Updates:

Switch to Active tab to see:

• Current progress percentage

• Current executing agent

• Completed subtasks

• Estimated time remaining

Switch to Live tab for immersive view:

• Full visualization

• Live logs

• Agent animations

• Real-time metrics

WebSocket Connection:

💡 Tooltip: "Live connection for real-time updates. Green = connected, Red = disconnected"

• Shows connection status icon

• Auto-reconnects if dropped

• Updates every 1-2 seconds

---

Common Tasks

Task 1: Create Simple Security Review Workflow

Goal: Review code for security issues

Steps:

1. Click "Create Workflow"

2. Name: "Security Review - Main Branch"

3. Goal: "Review authentication module for SQL injection, XSS, and authentication bypass vulnerabilities"

4. Context:

   {
     "codegraph_project": "myapp",
     "file_path": "auth/middleware.py"
   }

5. Auto-select agents: ✅ Enabled

6. Click "Create & Execute"

⏱️ Time: 3 minutes setup + 2-4 minutes execution 🎯 Result: Security audit report

Task 2: Using a Template for PR Review

Goal: Automatically review pull request

Steps:

1. Go to Templates tab

2. Find "PR Review Workflow"

3. Click "Use Template"

4. Name: "PR #456 Review"

5. Context:

   {
     "repository": "https://github.com/myorg/myapp",
     "pr_number": 456,
     "branch": "feature/new-auth"
   }

6. Click "Create & Execute"

7. Go to Live tab to watch

⏱️ Time: 2 minutes setup + 3-6 minutes execution 🎯 Result: PR review with suggestions

Task 3: Monitoring a Running Workflow

Goal: Watch workflow progress and intervene if needed

Steps:

1. Go to Active tab

2. Find your running workflow

3. Note current stage and progress

4. Click "Watch" to open Execution Theater

5. Monitor:

   • Which agent is active

   • What task they're executing

   • Live logs for details

6. If Issues:

   • Pause workflow (⏸️ button)

   • Review logs

   • Stop if necessary (⏹️ button)

⏱️ Time: Continuous monitoring 🎯 Result: Awareness of execution state

Task 4: Analyzing Workflow Performance

Goal: Understand why a workflow succeeded or failed

Steps:

1. Go to History tab

2. Find the workflow execution

3. Click to open details

4. Review:

   • Timeline: Which stage took longest?

   • Agent Performance: Did any agent struggle?

   • Quality Scores: Which subtasks scored low?

   • Logs: Any errors or warnings?

5. Identify improvements:

   • Better agent selection?

   • More context needed?

   • Different execution policy?

⏱️ Time: 5-10 minutes 🎯 Result: Actionable optimization insights

Task 5: Pausing and Resuming a Workflow

Goal: Temporarily halt execution and resume later

Steps to Pause:

1. Go to Active tab

2. Find running workflow

3. Click ⏸️ Pause button

4. Confirm pause action

5. Workflow stops after current subtask completes

Steps to Resume:

1. Paused workflow shows "Paused" badge

2. Click ▶️ Resume button

3. Workflow continues from where it stopped

4. All context preserved

⏱️ Time: Instant pause, resume within seconds 🎯 Use case: Need to update context, wait for approvals, etc.

---

Advanced Features

Custom Workflow Templates

🔧 Advanced

Create reusable templates from your successful workflows.

How to Create Template:

1. Go to History tab

2. Find successful execution

3. Click "⋮" menu → "Save as Template"

4. Template Name: "My Custom Review"

5. Make Public: Share with team (optional)

6. Click "Save Template"

Template appears in Templates tab for reuse.

Workflow Orchestration Policies

🔧 Advanced

Fine-tune how the orchestrator manages execution.

Agent Selection Policy:

💡 Tooltip: "Algorithm for choosing agents. Default uses LLM reasoning."

• LLM-Driven: Orchestrator reasons through selection (recommended)

• Algorithmic: Fixed algorithm (fast, less flexible)

• Manual: You specify all agents

• Hybrid: Mix of manual + automatic

Task Decomposition Strategy:

💡 Tooltip: "How complex goals are broken into subtasks"

• Hierarchical: Nested subtasks with dependencies

• Flat: All subtasks at same level

• Adaptive: Adjusts based on complexity

Result Aggregation Method:

💡 Tooltip: "How multiple agent outputs are combined"

• LLM Synthesis: LLM combines results (highest quality)

• Concatenation: Simple joining (fastest)

• Voting: Consensus-based (for decisions)

• Weighted: Based on agent confidence scores

Workflow Variables and Context

🔧 Advanced

Pass dynamic data to workflows:

Context Variables:

{
  "repository": "{{repository_url}}",
  "branch": "{{branch_name}}",
  "pr_number": "{{pr_id}}",
  "reviewer": "{{github_username}}"
}

Variable Substitution:

• Use {{variable_name}} syntax

• Replaced at execution time

• Can reference environment variables

• Useful for templatized workflows

Conditional Execution:

{
  "condition": "{{quality_score}} < 0.7",
  "then": "run_additional_review",
  "else": "mark_approved"
}

---

Tips & Best Practices

Workflow Design

💡 Best Practices:

1. Start simple: Single agent workflows first

2. Clear goals: Specific, measurable objectives

3. Good context: Provide relevant information

4. Test incrementally: Don't create huge workflows initially

5. Use templates: Leverage proven patterns

Naming Conventions

Good workflow names:

• ✅ "Security Audit - Authentication Module - Q4"

• ✅ "PR Review - Feature/NewAuth - #456"

• ✅ "Data Analysis - Customer Churn - Weekly"

Poor workflow names:

• ❌ "Workflow 1"

• ❌ "Test"

• ❌ "New Workflow"

Performance Optimization

Make workflows faster:

1. Use parallel execution for independent tasks

2. Choose faster models for simple subtasks

3. Enable caching for repeated operations

4. Limit agent count (diminishing returns after 5-7 agents)

5. Provide good context (reduces back-and-forth)

Cost Management

Keep costs down:

1. Set budgets on workflow and agent level

2. Use cheaper models where appropriate

3. Monitor token usage in Monitoring tab

4. Enable caching to reuse results

5. Limit max tokens per agent response

💡 Tip: Check estimated cost before executing expensive workflows.

Quality Assurance

Ensure high quality results:

1. Use quality agents: GPT-4, Claude Opus for critical tasks

2. Provide examples: Show what good output looks like

3. Add validation steps: Have second agent review

4. Check quality scores: Investigate if <0.8

5. Iterate on failures: Learn from low-quality results

---

Troubleshooting

Workflow Stuck at "Queued"

Symptom: Workflow not starting, stays in queued status

Solutions:

1. Check agent availability:

   • Go to Agents page

   • Ensure required agents are Active

   • Check if agents at max concurrent tasks

2. Check resource limits:

   • System may be at capacity

   • Wait for other workflows to complete

   • Or increase system resources

3. Check workflow priority:

   • Low priority workflows queue longer

   • Increase priority if urgent

4. Check error logs:

   • Click workflow card

   • View logs for error messages

Workflow Failed During Execution

Symptom: Status shows "Failed" with error

Solutions:

1. View execution logs:

   • History tab → Find execution

   • Click to view details

   • Check logs for error

2. Common errors and fixes:

   "Agent timeout":

   • Increase timeout in workflow policy

   • Or simplify task complexity

   "Insufficient context":

   • Provide more context in workflow input

   • Upload relevant documents

   • Enable CodeGraph for code tasks

   "Tool call failed":

   • Check tool credentials

   • Verify tool is active

   • Test tool connection in Tools page

   "Token limit exceeded":

   • Reduce max tokens per agent

   • Use cheaper model

   • Simplify goal/tasks

3. Retry workflow:

   • Click "Re-run" button

   • Modify context if needed

   • Execute again

Slow Workflow Execution

Symptom: Taking much longer than estimated

Solutions:

1. Check current stage:

   • Live tab shows which stage

   • Some stages (decomposition, aggregation) can be slow

   • Normal for complex goals

2. Identify bottleneck:

   • Monitoring tab → Tools

   • Find slow tool calls

   • Consider removing slow tools

3. Optimize model selection:

   • Some models are slower (Claude Opus vs GPT-4)

   • Use faster models for non-critical subtasks

   • Check agent configurations

4. Check parallel execution:

   • Ensure policy allows parallelism

   • Independent tasks should run simultaneously

   • Sequential execution is slower

No Results Generated

Symptom: Workflow completes but no output

Solutions:

1. Check subtask results:

   • History → Execution details

   • View each agent's output

   • Results may be there but not aggregated

2. Check aggregation settings:

   • Workflow policy may have issue

   • Try different aggregation method

3. Check agent outputs:

   • Agents may have returned empty results

   • Review agent logs

   • Verify agents had necessary context

---

Integration with Other Features

Workflows + CodeGraph

Use CodeGraph context in workflows:

{
  "goal": "Review this code for issues",
  "context": {
    "codegraph_project": "myapp",
    "file_path": "services/auth_service.py"
  }
}

See CodeGraph Guide →

Workflows + RAG

Enable document search during workflow:

{
  "goal": "Generate deployment guide",
  "context": {
    "search_documents": true,
    "knowledge_categories": ["deployment", "infrastructure"]
  }
}

See RAG Guide →

Workflows + Tools

Agents automatically use tools they're assigned:

• GitHub tool → Can read repos, create PRs

• Slack tool → Can send notifications

• AWS tool → Can manage infrastructure

See Tools Guide →

---

Keyboard Shortcuts

• Ctrl/Cmd + N: New workflow

• Ctrl/Cmd + R: Refresh active workflows

• Ctrl/Cmd + L: Go to Live tab

• Ctrl/Cmd + H: Go to History tab

• Space: Pause/Resume selected workflow

• Esc: Close modals, exit full-screen

---

FAQ

How long do workflows take?

Depends on complexity:

• Simple (1-2 agents, basic tasks): 30 seconds - 2 minutes

• Medium (3-5 agents, moderate tasks): 2-5 minutes

• Complex (5+ agents, complex tasks): 5-15 minutes

• Very Complex: 15-30 minutes

Check estimated duration before executing.

Can I stop a workflow mid-execution?

Yes! Two options:

1. Pause: Temporary halt, can resume later

2. Stop: Permanent cancel, cannot resume

⚠️ Note: Current subtask completes before stopping.

What happens if an agent fails?

Depends on error handling policy:

• Stop on Error: Entire workflow fails

• Continue: Skip failed task, continue with rest

• Retry: Auto-retry (configurable attempts)

Failure is logged for analysis.

Can I use the same agents in multiple workflows?

Yes! Agents can handle multiple workflows simultaneously up to their max_concurrent_tasks limit.

The system will queue workflows if agents are at capacity.

How do I know which agents will be used?

For auto-select workflows:

• Not determined until execution

• Based on task requirements

• Shown during initialization phase

For manual-select workflows:

• You specify agents upfront

• Shown in workflow definition

View assigned agents in Active tab once workflow starts.

---

Next: Knowledge Base Guide →

Learn to manage documents, code, and multimodal content

---

API Reference

Build a Workflow

1. New Workflow → fill Name/Description/Policy.

2. Open Canvas → add Agent, Tool, Branch nodes; connect edges.

3. Save.

API

Authentication All API calls require headers:

X-API-Key: <your_key>
Authorization: Bearer <your_token>

• POST /api/workflows

{ "name":"Bug Triage","policy_id":"support_policy","graph":{"nodes":[...],"edges":[...]} }

Run & Monitor

Click Run, pass inputs (JSON), watch spans in Runs.

API

• POST /api/workflows/{id}/run → {"run_id":"run_123"}

• GET /api/runs?workflow_id={id}&limit=50

Promote a Run

Convert a successful trace into a workflow template.

API (optional) POST /api/workflows/promote → {"workflow_id":"wf_new"}

Diagram