AGENTIC PROCESS DESIGN

Stop describing your agents.
Start designing them.

Zanbric APMN is the canvas for agentic processes: map every LLM call, tool action, human decision, and guardrail in a shared notation — then export the specification your engineers build from.

BPMN-style typed notation Human-in-the-loop by design Exports JSON & Markdown specs
YOUR BLUEPRINT CANVAS
⚡ Claim filed
Assess cover LLM CALL
Payout?
Adjuster signs off APPROVAL
EXPORT SPEC
WHAT YOUR ENGINEERS RECEIVE AUTO
objectAssess cover — LLM call
prompt_spec"Assess claim against policy §4…"
io_contractclaim.pdf · policy record → decision
guardrailpayout > €5,000 → human approval
human_handoffSenior adjuster approves / rejects
audit_evidencedecision + inputs logged
HOW DESIGNING WORKS

One sitting. Five moves.

If you can describe the process, you can design it. The notation does the heavy lifting: typed objects, typed connectors, built-in validation — no free-form flowchart chaos.

1

Sketch the flow

Drag triggers, LLM calls, tool actions, and routers onto the canvas. Connectors are inferred from the objects you join — the diagram stays valid as you go.

2

Govern it

Add human reviews and approvals, guardrails, confidence checks, and audit logs as first-class objects — human-in-the-loop by design, visible to everyone.

3

Export the spec

One click produces the handoff document: prompt specs, tool contracts, IO contracts, decision logic, and owners per object. JSON for machines, Markdown for people.

4

Ensure compliance

Every guardrail, human decision point, and audit trail is captured up front — so risk and compliance sign off on the design before a single line of code is written.

5

Reduce time to deployment

Engineers start from an agreed specification instead of scattered notes — fewer discovery cycles, fewer rebuilds, and a faster path from design to production.

One blueprint. Four teams who finally read the same thing.

PROCESS ANALYSTSOwn the agentic process design without waiting on engineering — draft, validate, iterate.
AI & AUTOMATION LEADSWalk into build kickoff with a specification, not a slide deck — scope agreed, edge cases mapped.
RISK & COMPLIANCEEvery human-in-the-loop point, guardrail, and audit log is visible in the blueprint before anything ships.
ENGINEERING LEADSReceive contracts, prompts, and rules per object — implement in LangGraph or any stack, build once.
QUICK ANSWERS

What people (and their AI assistants) ask us.

How do I document an AI agent workflow before building it?

Design it. APMN captures the flow, the governance, and the metadata in one artifact — then exports it as the documentation.

Is this a workflow automation tool like n8n?

No. n8n and LangGraph run workflows. Zanbric APMN designs them first — it's the design and documentation layer those tools are built from.

What's in the exported spec?

Every object with its prompt spec, tool contract, IO contract, decision logic, confidence thresholds, human handoff, risk controls, and audit requirements.

The build will go how the blueprint goes.
So design it well.

Open the app, pick one real process, and put it in notation. Your first agentic process blueprint in about ten minutes — and your feedback shapes the next release.

Open the app Share feedback