Agentic Process Automation. Out of the box.
Purple8 gives AI agents everything they need to run real business processes end-to-end: stateful workflows, SLA enforcement, human-approval gates, knowledge graph memory, and a full audit trail. Through 82 MCP tools, in a single process. An agent that can call journey.advance can run a loan application, a compliance review, or a supplier approval from start to finish.
What makes it agentic
Not RPA. Not BPM. Agentic.
- •Records UI clicks — breaks when screens change
- •No reasoning — fixed scripts only
- •No memory between runs
- •Fails silently on exceptions
- •Fixed flowcharts — every path pre-defined
- •Requires developers to model every exception
- •Audit trail in a separate log system
- •Separate orchestration service to operate
- •Agent reasons about intent and context
- •Handles exceptions without pre-programming
- •Knowledge graph memory persists across runs
- •SLA enforcement and HITL built in
- •Audit trail is the graph — immutable, always on
How it works
Five steps. Zero orchestration service.
Define the process
Call journey.define with stage definitions, SLA policies, and HITL requirements. Natural language descriptions are enough.
Agent starts an instance
journey.start creates a workflow instance stored as a graph node. Every state transition becomes an edge — the graph IS the process state machine.
Agent advances through stages
The agent calls journey.advance after completing each stage's work. SLA timers run automatically; breaches create SLA_BREACHED edges.
Humans approve where required
HITL gates pause the workflow. A portal surfaces pending decisions via journey.list_hitl. The approver's decision is an immutable graph edge.
Audit trail is always there
Every transition, AI action, and human decision is stored as a graph edge. journey.audit returns the complete, tamper-evident history of any instance.
Use cases
What teams are automating
Loan origination & compliance
An agent ingests the application, extracts entities via DocIntel, traverses the customer's relationship graph to assess risk, advances through underwriting stages, gates on human review, and maintains a full audit trail — without application code.
Patient pathway management
Multi-stage clinical workflows with SLA enforcement per stage. Agents track referrals, flag delays, surface HITL tasks for clinicians, and log every state transition immutably.
Procurement & supplier approval
Agents route purchase requests, verify supplier credentials against the knowledge graph, enforce spend limits, escalate exceptions to the right approver, and close the loop with automated PO generation.
Security incident response
SOC vertical detects anomalies, triggers a containment workflow, escalates to human analysts for critical decisions, and documents the entire response chain as an auditable graph.
Contract lifecycle automation
Ingest contracts via DocIntel, extract obligations and parties, track milestone dates through Journey Engine stages, alert when SLAs approach, route renewals to stakeholders with full version history.
Customer onboarding
KYC document ingestion, entity extraction, risk scoring via graph traversal, staged approval with HITL gates, and automated account provisioning — end-to-end, agent-driven.
Stack consolidation
One process replaces five services
FAQ
Common questions
What is agentic process automation?
Agentic process automation (APA) is the use of AI agents — rather than humans or rigid scripts — to orchestrate, execute, and adapt multi-step business processes end-to-end. Unlike traditional RPA (which records mouse clicks) or BPM (which follows fixed flowcharts), agentic automation reasons about context, handles exceptions autonomously, and learns from each run.
How does Purple8 support agentic process automation?
Purple8's Journey Engine provides the stateful workflow layer AI agents need: multi-stage process definitions, SLA enforcement, human-in-the-loop approval gates, and an immutable audit trail — all stored in the graph and accessible to agents through 82 MCP tools. An agent can start a process instance, advance it through stages, handle exceptions, and surface decisions for human review entirely through natural language tool calls.
What is the difference between agentic automation and RPA?
RPA automates fixed sequences of UI actions and breaks when screens change. Agentic automation uses AI agents that reason about intent, retrieve context from a knowledge graph, adapt to new inputs, and make decisions — without being programmed for every edge case. Purple8 provides the persistent memory, workflow state, and tool interface that make agentic automation reliable in production.
Does agentic automation require a separate workflow engine?
Not with Purple8. The Journey Engine is built into the same process as the graph database, vector search, RAG pipeline, and authentication layer. There is no Airflow, no Temporal, no LangGraph — agents call journey.* MCP tools directly and the workflow runs in-process.
Can Purple8 handle human-in-the-loop (HITL) steps in agentic workflows?
Yes. Stages in a Journey Engine process can require human approval before the instance advances. The journey.list_hitl MCP tool surfaces pending decisions; journey.resolve_hitl unblocks them. The audit trail of who approved what, when, and why is stored immutably in the graph as edges.
Ready to automate your first process?
Free to start, runs on your machine, and takes about 60 seconds to install.