Visual core steps
Configure prompt, retrieval, model, approval and merge steps in the visual workflow experience.
AI workflow orchestration
Build AI workflows around the way your organisation works
Pūnaha connects prompts, knowledge retrieval, supported AI models, parallel stages, approvals and merged results. Workflows can be validated, versioned, published and followed through execution.
Knowledge
Models + decisions
Approved outcome
A flexible sequence
Connect the right building block at each stage
This example shows a common pattern. A workflow can be simpler or include parallel branches and merged results.
- 1
Prompt
Shape the task and the information it needs.
- 2
Retrieval
Bring in relevant context from an approved knowledge store.
- 3
Model
Use the configured local or external deployment for this step.
- 4
Approval
Pause for a person to review, approve or reject.
- 5
Merge
Combine results and continue with a traceable execution record.
Workflow capabilities
Start visually, extend through configuration
The current visual workflow experience focuses on the most useful core steps. Additional backend nodes can be configured by developers.
Parallel stages
Run suitable work in parallel and bring the results together later in the workflow.
Conditions
Use backend workflow configuration to choose a path based on process data.
REST connections
Use generic REST steps in developer-configured workflows to connect approved HTTP services.
Read-only MySQL
Use a developer-configured read-only MySQL step when a workflow needs approved database information.
Web search backend
Configure a generic web search backend for use cases that require external search.
Controlled change
Treat workflows as operational assets
Design
Arrange the steps, models, knowledge and decisions for the intended process.
Validate
Check the workflow definition and required connections before publication.
Version and publish
Publish a defined version rather than changing a live process without a clear boundary.
Execute and inspect
Run the workflow durably and review its execution history and approval state.
Human in the loop
Use automation where it helps. Stop where judgement matters.
A workflow does not have to choose between full automation and manual work.
Approval steps let a workflow pause while a person reviews the available context and proposed result.
This can support processes where policy, accountability or consequences make an automatic decision inappropriate. The implementation still needs to define who can approve and what happens next.
Review
Show the relevant material and proposed result before continuing.
Approve or reject
Capture the person's decision as part of the workflow state.
Continue deliberately
Follow the configured path after the decision is recorded.
AI workflow questions
Is Pūnaha an AI agent orchestration platform?
Pūnaha orchestrates model calls, retrieval, decisions and actions in governed workflows. It can support agent-like processes, but this site uses precise workflow language instead of suggesting unchecked autonomous behaviour.
Can a workflow connect other AI products?
Yes. Pūnaha supports configured local models and several external model providers. Generic REST steps can also be developer-configured for approved HTTP services.
Can workflows include a person?
Yes. Human approval is a first-class workflow step and can pause execution until a decision is recorded.
Are workflows versioned?
Yes. Workflow definitions can be validated, versioned and published, with durable execution and execution history.
Bring a process you want to make clearer
We can use a private demonstration to map its knowledge, model, decision and approval steps.