Bangun dengan AI agent Anda
Build a NocoBase app — AI Workflows: native workflows with AI-employee nodes and an execution flow. Match the layout and signature visuals of this reference prototype: https://static-docs.nocobase.com/solution/templates/11-ai-workflow-automation.html
Sebelum memulai, ikuti panduan cepat AI agent untuk menginstal NocoBase dan menghubungkan agent Anda. Hasil AI dapat bervariasi; bergantung pada model dan kompleksitas sistem, mungkin diperlukan penyesuaian atau beberapa putaran tambahan.
Introduction
Use your favorite AI agent and NocoBase to quickly build a customizable, reliable, and continuously evolvable AI workflow automation system for managing business trigger events, condition checks, AI processing, data updates, execution results, and workflow run history.
You can copy the prompt below and let your AI agent generate the basic structure of the AI workflow automation system in NocoBase, then visually adjust triggers, conditions, AI nodes, and follow-up actions.
This system fits scenarios such as inventory suggestions, ticket classification, contract clause analysis, knowledge-base summarization, content publishing, asset synchronization, employee onboarding suggestions, and any case where AI should take part in business processes automatically.
AI workflow list:
Workflow execution results:
Visual workflow canvas:
What problems does an AI workflow automation system solve?
Many teams already use AI to generate content, summarize information, or provide suggestions, but these actions usually stay inside one-off conversations. Users have to manually copy data, call the AI, check the results, and write the output back into the business system — the whole process still depends on manual work.
With an AI workflow automation system, AI can participate directly in your existing business processes. For example, when an inventory record changes, a restocking suggestion is generated automatically; when a knowledge-base article is created, a summary is generated automatically; when a ticket is submitted, its type is classified automatically; when a contract enters the approval stage, clause changes are summarized automatically.
Each workflow can be triggered by a business event and execute calculation, condition checks, LLM, AI employee, record update, or record creation steps in the configured order. The AI’s input, instructions, output, and final write-back status are all kept in the execution records.
This way, AI is no longer just a standalone chat tool — it becomes an automated execution node inside the business process. Teams can clearly see why each workflow was triggered, what the AI processed, what results it produced, and whether the data was updated successfully.
Core features
AI workflow management
- Centralized workflow management: View workflow names, trigger types, enabled status, and total execution counts in one place.
- Enable and disable control: Quickly turn specific workflows on or off as the business requires.
- Runtime overview: Track the number of enabled workflows, AI workflows, and accumulated executions to understand automation usage.
Business events and execution modes
- Data event triggers: Start workflows automatically when records are created, updated, or other collection events occur.
- Manual execution: Run a specific flow by hand for testing or for processing particular business data.
- Asynchronous runs: Move long-running AI tasks to the background to reduce the impact on the user’s current operation.
Visual flow orchestration
- Node canvas: Connect triggers, processing nodes, and end nodes on a visual canvas to see the complete flow at a glance.
- Conditional branches: Decide whether to call AI — and which follow-up action to run — based on field values or business rules.
- Execution order configuration: Flexibly combine calculation, condition, AI processing, and data operation nodes.
AI and data processing nodes
- LLM nodes: Handle well-defined tasks such as content summarization, classification, information extraction, and text generation.
- AI employee nodes: Complete more comprehensive business processing based on preset roles, prompts, and tools.
- Data operation nodes: Create records, update fields, and write AI-generated results back into business data.
Execution results and history tracking
- Full execution details: Show trigger input, AI instructions, model output, node statuses, and the final write-back result.
- Execution history filtering: Review each run’s time, input parameters, and status per workflow.
- Problem diagnosis: Use node-level inputs and outputs to determine whether a failure happened in the condition, the AI call, or the data update.
Testing and ongoing maintenance
- Manual test runs: Execute a workflow before enabling it to verify conditions, AI output, and data write-back.
- Workflow versioning: Keep different configuration versions for continuous adjustment and rollback.
- Flexible node extension: Add new conditions, AI capabilities, and follow-up actions as the business evolves, without rebuilding the whole flow.
Why build a workflow automation system with AI and NocoBase?
If you build from scratch with plain vibe coding, every new AI automation scenario usually means reimplementing event listening, data queries, model calls, condition logic, error handling, and result write-back. As flows multiply, the code grows more complex and executions become hard to trace.
NocoBase already provides the foundational capabilities of a business system — data models, workflows, condition checks, record updates, permissions, and operation history. You can drop LLM or AI employee nodes directly into these capabilities, letting AI work with real business data and processes instead of sitting outside the system.
For example, after an inventory record is created, an inventory suggestion can be generated automatically; after a knowledge-base article is added, a summary can be generated automatically; after a contract’s status changes, AI can summarize the stage change and then create an audit record. All of these flows can be viewed and adjusted on the visual canvas.
The input, AI instructions, output, and data update status of every execution are recorded, making it easy for teams to review results and locate problems. For teams that value traceability, maintainability, and long-term iteration, NocoBase is the better choice for building AI workflow automation that goes into real business use.
FAQ
- How is AI workflow automation different from regular workflows?
Regular workflows mainly execute predetermined rules, such as condition checks, sending notifications, creating records, or updating fields.
AI workflows can add LLM or AI employee nodes into the flow, letting the system perform summarization, classification, information extraction, content generation, and recommendation based on business context. AI handles the unstructured tasks, while the workflow controls trigger conditions, execution order, and result write-back.
- How do AI workflows differ from automation scripts generated by vibe coding?
Vibe coding can quickly produce an automation script, but as business scenarios grow, event triggering, model calls, data permissions, error handling, logging, and version maintenance all require repeated development.
NocoBase places AI nodes inside its existing data model and workflow system. Teams can configure triggers, conditional branches, AI processing, and data operations on a visual canvas and review every execution — better suited to managing continuously running business automation than maintaining scattered one-off scripts.
- Which business events can trigger AI workflows?
AI workflows can be triggered automatically by collection events such as record creation or updates, or executed manually by users.
For example, generate a restocking suggestion after an inventory record is updated, classify the issue type after a ticket is created, generate a summary after a knowledge-base article is added, or analyze clause changes automatically after a contract enters a specific stage.
- Can I set conditions before calling AI?
Yes. You can first check whether the data meets specified conditions before entering an LLM or AI employee node.
For example, generate a restocking suggestion only when inventory drops below the safety level, call AI analysis only for high-priority tickets, or create an audit record only after a contract enters the approval stage. This reduces unnecessary model calls and keeps the flow aligned with real business rules.
- Can AI-generated results be written back into the business system directly?
Yes. After AI finishes processing, the flow can continue with record update or record creation nodes, saving summaries, classifications, suggestions, tags, or analysis results into the corresponding tables.
For example, write inventory suggestions back to inventory records, save ticket classification results to ticket fields, or create a new audit record based on contract analysis results.
- Can I see which inputs the AI used and what results it generated each time?
Yes. Execution details show the trigger input, the instructions used by AI nodes, the model output, subsequent node results, and the final data write-back status.
This means teams can see not only whether a flow succeeded, but also why the AI produced a particular result — useful for validating output, refining prompts, and troubleshooting anomalies.
- How do I locate problems when an AI workflow fails?
Every run keeps its execution time, node statuses, input parameters, and output results. Teams can use the execution history to determine whether a failure happened in the trigger condition, calculation, condition check, AI call, or data update step.
Compared with standalone scripts that only show a final error message, this node-level execution record is far better suited to debugging long-running business processes.
- Can I use LLM nodes and AI employees together?
Yes. LLM nodes are good for well-defined, single-purpose content tasks such as summarization, classification, extraction, or text generation.
AI employees can be configured with roles, prompts, and available tools, suited to tasks that combine business data with multiple operation steps. Teams can choose one based on flow complexity, or combine both across different workflows.
- Can Claude Code, Codex, Cursor, or OpenCode help build these workflows?
Yes. AI coding agents such as Claude Code, Codex, Cursor, and OpenCode can connect to NocoBase and generate data structures, workflow nodes, condition logic, and related pages from natural-language prompts.
After the workflows are generated, teams can still view, test, and adjust them in NocoBase’s visual interface instead of relying on AI to keep rewriting code.
- Can I control who can view, run, or modify AI workflows?
Yes. NocoBase can control by role which execution records users can view, which flows they can run manually, and whether they can modify workflow configuration.
For example, regular business users can only view results related to themselves, process administrators can review execution history and adjust nodes, and system administrators can manage model configuration, workflow versions, and permissions.
- Can AI workflows be iterated and maintained over the long term?
Yes. As business rules change, you can keep adjusting trigger conditions, prompts, models, branch logic, and follow-up actions, or add nodes and swap processing approaches.
Workflow versions, execution history, and node results help teams verify the effect of changes, avoiding the need to rebuild a whole automation script every time requirements shift.
- Are AI workflows suitable for real enterprise use?
Yes — especially for automation scenarios that need stable operation, traceable processes, and verifiable results.
NocoBase can combine permission management, workflow versioning, execution history, operation logs, audit logs, enterprise SSO, APIs, and plugin extensions as needed. Compared with one-off AI automation demos, it is far better suited to business processes that require ongoing maintenance and gradual expansion.
