Flowable Agentic Case Platform

Put AI agents to work. Keep the control your auditors expect.

Flowable orchestrates people, systems, and AI agents around your cases and processes, on open standards, so every step an agent takes is governed, explained, and yours.
PeopleSystemsEventsAI agentsCase
PeopleSystemsEventsAI agentsCase
  • Every enterprise is experimenting with AI agents right now.

    Mostly for personal productivity: coding agents, chat co-workers, drafts and summaries. Letting agents run the business itself is the next step, and there enterprises are still holding back. With good reason: agents work, most of the time.

  • Most of the time is not good enough.

    Not for a bank running a compliance check. Not for an insurer settling a claim. Not for a hospital scheduling care. When an agent acts on real customers and real money, "usually right" is a liability.

  • The intelligence behind an agent is rented.

    Everyone rents the same models, on the same terms. What makes your business yours is the context the agent works in: your processes, your data, your rules, your history, and who is allowed to do what. That knowledge of how you run your business is your intellectual property. Hand it to an AI provider and you hand over what makes your company different. Owning that context, and keeping agents inside it, is called orchestration.

Flowable is the orchestration layer for agentic AI.

How it works: the case is the harness

Step 1

Start with something every business knows: a case.

A customer onboarding. An insurance claim. A loan application. A case has stages, rules about what may happen when, and a clear goal. Businesses have run work this way forever.

Step 2

People work the case.

They review documents, make judgment calls, approve decisions. The case records who did what, and when.

Step 3

Systems do the predictable work.

Structured processes, data lookups, document generation, decisions by rulebook. If the same input always produces the same output, it runs deterministically, the same way every time, no AI needed.

Step 4

Now add AI agents.

They read incoming documents, check policies, draft assessments. Each agent sees only the data it needs for its task, and can only act where the case allows it. The same need-to-know discipline you apply to people, applied to agents with more precision. You can run every use case at this level with agents as the scoped workers inside a case you control.

Step 5

Want to go further? Add an agent for the orchestration itself.

If you want AI not just working the case but steering it, you add the orchestrator agent. Every time something changes, it looks at what the case allows right now, reasons about the best next step, and either takes it or suggests it to a person. High-stakes actions always wait for a human.

The case is the harness: it decides what an agent may do. The agent decides what to do next. You stay in control of both.

Watch a case run itself

Below is an example Know Your Business onboarding case. Step through it and watch AI agents, deterministic process steps, and people share the work.

Know Your Business onboardingNot started
Orchestrator Agent
Attached to this case. Every time something changes, it reevaluates what should happen next.
Receive application & documents
deterministic
Classify incoming documents
Document Agent
Extract structured data from documents
Document Agent
Verify against company registry
deterministic
Build business profile
Utility Agent
Case timeline
Run the case to see the audit trail fill in.

Want to run a case like this yourself?

Start your Flowable Trial

The operator view of the same case

What the business sees above, an operator can open in Flowable Hub: the orchestrator instance behind the case, with every invocation, prompt, tool call, guardrail verdict, token count, and cost. Select any span in the timeline to inspect it.

KYB Onboarding Orchestratoragent instanceCase CASE-2026-04182 · Meridian Trading GmbH
No agent activity yet

The operator view fills in as the case runs. Start the case above and every agent call, prompt, guardrail check, and cost will appear here as it happens.

One case, scoped context for every agent

The case holds the entire context of the run: every value an agent produced and every decision a person confirmed. No agent sees all of it. Hover over an agent to trace which fields it reads and which it is allowed to write.

Case data · CASE-2026-04182
agent readsagent writes
Application
ChannelPortal
Requested productBusiness current account
ReceivedJul 10, 2026, 09:02
Documents (4)
Registration extractHR-Auszug_Meridian.pdf · verified
Shareholder registerAktienregister_2026.pdf · verified
ID copyKeller_Pass.jpg · verified
Bank statementKontoauszug_Q2.pdf · verified
Company profile
Legal nameMeridian Trading GmbH
Registration numberCHE-234.567.891
DirectorsA. Keller, M. Fontana
Shareholders3
Ownership above 25%1 shareholder
Business profileActive trading company, 2 directors …
Screening
FindingPossible PEP match · M. Fontana · 0.71
Enhanced screeningClear · different person
Human clearanceConfirmed · compliance analyst
Policy check
CompliantYes
CitationsPOL-114, POL-201, POL-088
Product fit
RecommendationBusiness banking package
Decision
Risk classB · DMN riskClass v3
Assessment draftrisk-assessment.pdf · draft
Sign-offApproved · compliance officer
Agents on this case · tap to tracehover to trace

The pattern: the case owns the full context. Each agent gets a scoped slice in, and may only fill the fields its task owns.

Deterministic steps cost zero tokens. AI runs only where judgment is needed.

Four different models in one case. Swap any of them without touching the business logic.

Each agent saw only what it needed. The case scoped the context.

Every call, token, cent, and guardrail check is in the audit trail.

You saw one case. You'll run tens of thousands or even millions.

The operator view above inspects a single case. At scale you need the global picture: dashboards aggregate every agent invocation across all cases with volumes, token consumption, cost, durations, tool calls, guardrail violations. All broken down by agent and by model. Operators spot drift, spikes, and runaway costs before they matter.

Agent operations dashboardall agents · all cases · hover any chart
533
Agent invocations
7
Unique agents invoked
1,625,267
Total tokens consumed
7,956.89 ms
Average invocation duration
834
Total tool calls
55
Guardrail violations
Agent invocations over time
39236May 5May 11May 17May 23
Agent invocations
Total token usage over time
88.2K53.0K17.9KMay 5May 11May 17May 23
Total tokens
Agent invocation average duration
9.4K7.8K6.2KMay 5May 11May 17May 23
Avg duration (ms)
Agent invocations by definition
kybOnboardingOrchestratorkybGuardrailAgentkybDocumentAgentkybScreeningAgentkybPolicyKnowledgeAgentkybProductFitAgentkybRiskAssessmentAgent
LLM distribution
LLM distribution
533
Invocations
claude-opus-4-8gpt-5.2-codexclaude-sonnet-4-6gpt-4ogemini-2.5-pro
Input vs output tokens by LLM
claude-opus-4-8gpt-5.2-codexclaude-sonnet-4-6gpt-4ogemini-2.5-pro
Input tokensOutput tokens

Under the hood

Everything above is built from a small set of governed building blocks. Expand what interests you.

Eight kinds of agents

Built-in control at every step

Not just agents. A full case platform underneath.

Everything on this page runs on the deterministic capabilities Flowable has run enterprise operations on for years. The agents are the newest participant.

Processes & cases

BPMN processes and CMMN cases: structured flows and long-running, plan-driven work.

Rules & decisions

DMN decision tables for the calls that must be deterministic.

Integrations

REST and service connectors to your core systems, modeled and governed like everything else.

Events & messaging

Asynchronous events over message brokers start and steer cases as the outside world changes.

Human work

Task management, forms, and role-based assignment for the people in the loop.

Documents & content

Case documents, templates, and generated output managed alongside the work itself.

The Agentic Case Platform is the engine that enterprises trust for their core processes: AI agents, open standards, durable state, transactions, and error handling for work that can run for years. The visual model your experts review is exactly what executes. There is no drift between the diagram and the deployment, which makes the model itself the auditable artifact, for humans and machines alike.

Do not burn tokens on deterministic work. If the same input always produces the same output, model it, do not prompt it.

Rent the intelligence, own the context. The model is rented by everyone on the same terms. Your data, rules, and history are what make it yours.

Codify business knowledge into human-readable, auditable artifacts. An automated process is a valuable asset. Treat it like one.

Open standards: BPMN, CMMN, DMN

Durable execution: state, transactions, retries built in

Every model versioned, permissioned, audited

Model-neutral: swap the LLM, keep the logic