AI and knowledge
Configure prompts, semantic or hybrid retrieval, local or external model calls and durable knowledge writes.
AI workflow and process orchestration
See the whole job around the AI
Most useful automation is not one prompt. It starts somewhere, gathers information, makes choices and sometimes waits for a person. Pūnaha lets you lay out that work as a visual workflow, so people can see what happens and change it without losing the history.
A flexible sequence
Start where the work starts
A workflow might begin with a question, a file, a schedule or an event in another system. From there it can take a short route, split into parallel work or join a wider BPMN-based process.
- 1
Trigger
Start from input, a webhook, a schedule or a supported process event.
- 2
Ground
Retrieve approved knowledge, validate data and prepare the task.
- 3
Work
Use models, rules, documents, connections or a reusable child workflow.
- 4
Review
Pause for approval or an assigned human task when judgement is required.
- 5
Continue
Publish a message, update knowledge, return an output or move to the next process stage.
Visual workflow modules
A workflow can do more than pass prompts between models
The visual builder brings the supported modules into one place. A developer can connect knowledge, decisions, documents, integrations and AI without hand-editing the core workflow definition.
Logic and decisions
Transform and validate data, evaluate conditions and decision tables, merge branches and record approval outcomes.
People and governance
Assign human tasks, set due and overdue behaviour, pause for approval and add digest-backed evidence checkpoints.
Documents
Extract or split text, apply configured redaction and classification rules, render templates and create standard file outputs.
Connections
Use approved HTTP, email and database connections, plus a durable local message queue for publish, receive, retry and dead-letter handling.
Explore connectionsFlow control
Use persistent waits, correlated events, bounded batch processing, parallel paths and pinned or latest child-workflow versions.
Operational execution
The work should not disappear when someone closes a browser
Schedules
Start a published workflow once, at an interval or from a cron schedule with an explicit time zone and bounded catch-up window.
Webhooks and events
Begin or resume work from approved external requests and correlated events.
Durable messages
Hand work to competing consumers, lease a message and acknowledge, retry or dead-letter it explicitly.
Persistent waits
Retain timers, approvals, tasks and events without relying on a continuously open browser.
Restart-aware state
Persist workflow versions, runs, waits, events, outputs and worker leases in the Pūnaha platform.
Inspect execution
Review the current state and append-only execution events when diagnosing or governing a run.
Controlled change
Change a live workflow without losing track of the version
Design
Arrange the triggers, data, knowledge, models, rules, integrations and people required for the intended work.
Validate
Check the workflow structure, module configuration and referenced resources before publication.
Version and publish
Publish an immutable version rather than changing a live process without a clear boundary.
Execute and inspect
Run the workflow durably and review events, waits, human work, outputs and failure state.
Structured process automation
Use a process model when the route needs to be explicit
For structured processes, Pūnaha supports BPMN-based activities, events, gateways, subprocesses, transactions, data and diagram interchange.
Use the provider-neutral workflow modules for product capabilities and the BPMN-based process model for the standard process elements that coordinate them.
Pūnaha does not currently make a formal BPMN conformance claim. Independent review and external interoperability evidence remain separate release requirements.
Processes
Model repeatable work with explicit events, activities, gateways and responsibilities.
Explore BPMN automationDecisions
Keep repeatable rules explicit through conditions and decision tables.
Cases
Use CMMN-based adaptive case management when the next appropriate task depends on evidence and judgement.
Explore adaptive casesAI workflow questions
Is Pūnaha an AI agent orchestration platform?
Pūnaha coordinates model calls, retrieval, decisions, tools and human work inside a defined workflow. It can support agent-like processes, but we use precise workflow language rather than suggest unchecked autonomous behaviour.
Can a workflow connect other AI products?
Yes. Workflows can use configured local models and supported external model providers. Approved HTTP connections can also connect services that expose suitable APIs.
Which modules are available visually?
The visual builder covers triggers, prompts, retrieval, models, data transformation and validation, conditions, decision tables, approvals, human tasks, documents, HTTP, email, databases, message queues, waits, events, batches, child workflows, evidence checkpoints, merges and outputs, together with supported BPMN elements.
Can a workflow continue after a person closes the browser?
Yes. Supported waits, approvals, human tasks, events and run state are persisted by the Pūnaha platform and can resume independently of the original browser session.
Are workflows versioned?
Yes. Definitions are validated and published as immutable versions, with durable execution records and support for pinned or latest child-workflow versions.
Show us the process people currently hold together by hand
We can use a private demonstration to map where it starts, what information it needs, who makes the decisions and what should happen when something goes wrong.