Step 1
Receive and shape the task
A prompt step turns the request into the defined inputs needed by the workflow.
What you can see
The published workflow version defines the expected structure.
Product tour 03
Use a different AI model when the next step needs one
A workflow does not have to depend on one model. This tour sends different parts of a task to different configured deployments, brings the results together and asks a person to decide whether the combined work should continue.
One task
Multiple model steps
Reviewed result
Interactive walkthrough
Five steps, with the model chosen for the job
Each model is configured for a particular step. The tour does not pretend that one provider is best at everything.
Step 2
Add trusted context
A retrieval step finds relevant organisational knowledge before any model is asked to work with it.
What you can see
The selected knowledge store remains separate from the model deployment.
Step 3
Run model steps in parallel
Independent parts of the task can use different supported local or external model deployments at the same time.
What you can see
Each step refers to its own configured deployment.
Step 4
Merge the outputs
A merge step brings the branch results into the form needed for the next decision.
What you can see
The workflow defines how branch outputs continue.
Step 5
Review before continuing
A person reviews the combined material and decides whether the workflow should proceed.
What you can see
Approval adds a visible accountability point.
Model options
Choose deployments around the work and the data
| Option | How Pūnaha connects | Question to evaluate |
|---|---|---|
| Local GGUF | Local execution through llama.cpp | Does this task need to stay with a local model? |
| OpenAI or compatible | Configured external API deployment | Do the service capabilities and data terms fit? |
| Azure OpenAI | Configured Azure deployment | Does it fit the organisation's Azure architecture? |
| Anthropic | Configured external model deployment | Is this model suitable for the step and environment? |
| Google Gemini | Configured external model deployment | Is this model suitable for the step and environment? |
Bring the model choice you are trying to untangle
We can show how the task, knowledge, deployments, branches and approval point fit together in one workflow.