Visual node canvas
A React Flow-powered designer with 16 node types — decision branching, parallel execution, ForEach loops, and conditional routing with full JSONLogic expressions.
AgentFlow · Workflow Automation
Workflows that think, decide, and act.
Design complex business processes on a visual canvas that blends AI agents with human judgment. From simple approval loops to multi-stage document workflows with CRM integrations — all self-hosted, with complete control over your data and execution.
16 node types · AI agents · Safe test-run mode
AI agents
HubSpot CRM
OIDC / SSO
PDF & DOCX
Webhooks & SQL
Capabilities
Everything a real workflow engine needs
A visual designer, production-grade AI agents, human-in-the-loop tasks, and enterprise integrations — in one coherent platform.
AI agents
Run AI agents inside workflows with structured JSON output, validation, retries, and configurable creativity — with knowledge/RAG and automatic source citations.
Human-in-the-loop
Route approvals with granular assignment — individual, role, manager-chain, or dynamic expressions — plus custom forms and read-only document review.
Documents & forms
Generate PDF/DOCX with a WYSIWYG editor, variable interpolation, headers/footers, and Vietnamese government format (Nghị Định 30). JSON Schema forms with PII marking.
Enterprise integrations
HubSpot CRM read/write with webhook triggers, generic webhook nodes with auth & retries, read-only SQL execution, and a provider-neutral OAuth connection framework.
Analytics & audit
A process-mining dashboard for cycle times and bottlenecks, a unified task inbox, immutable audit trails, and per-user notification channels with durable delivery.
AI agents, under control
Production-grade AI, not a demo
AgentFlow treats AI as a first-class, governed workflow step — with the guardrails enterprises actually need to run it in production.
- Per-node token budgets with real-time enforcement — exceed the limit and the workflow routes to a failure handler. No surprise bills.
- Knowledge & RAG — attach documents with OpenAI or Cohere embeddings; agents semantic-search and cite retrieved sources inline.
- Structured output with JSON Schema validation and automatic retries — agents return typed data your workflow can branch on.
- Deployment kill switch — disable all agent calls platform-wide in seconds, no redeploy, fully audited.
Agent · Extract contract terms
AI agent · budget 8k tokens
3 sources cited · validated ✓
4,180 / 8,000 tokens used
Simulation mode
Test with real agents, zero risk
The differentiator teams love: run unpublished workflows with real routing and real AI agents, but every external side-effect is safely captured instead of sent.
- Emails, webhooks, and CRM writes are captured for inspection — database writes are always blocked.
- Step through human tasks yourself, then rewind and re-run from any step — upstream results are reused.
- Flip individual steps to "really send", or arm a live capture to catch the next real inbound event.
- Watch execution node-by-node in near-real-time — reconnects re-sync without losing steps.
Start → Agent → Decision
completed · captured
Email · Notify approver
captured — not sent
Human · Step through as admin
waiting for you
Why AgentFlow
Built for teams that ship AI to production
Immutable snapshots
Every execution binds to a point-in-time definition snapshot. Rewind forks a new attempt rather than mutating history — true audit fidelity.
Per-node AI budgets
Token limits on individual agent nodes and per instance. Exceed them and the workflow routes to a handler — no runaway costs.
Redaction by default
Non-admins see redacted payloads — structured paths filtered, free-form text omitted — in both production and test-run data.
Fine-grained RBAC
17 permissions across design, publish, start, tasks, instances, integrations, and audit — assignable to system or custom roles, per tenant.
Use cases
Automation for every team
AgentFlow orchestrates decisions, approvals, and system integrations across the business — the same engine works whether or not a document is involved.
Lead qualification & routing
An AI agent scores and enriches each inbound lead, dedupes it against your CRM, routes hot ones to the right rep, and books the follow-up — deals over a threshold branch to deal-desk approval.
Access requests & provisioning
Joiner-mover-leaver requests: the manager approves, an agent checks segregation-of-duties policy, accounts are provisioned via webhook, and every step lands in the audit log.
Ticket triage & escalation
AI classifies inbound tickets, sets priority, auto-resolves the routine ones, routes the rest to the right queue, and escalates SLA breaches to a manager — no document in sight.
Purchase & expense approvals
Requisitions route through amount-based approval chains, issue the PO to your ERP via webhook, and notify the vendor — with AI risk checks on new suppliers.
Claims & loan decisioning
An AI agent pre-screens each claim or application, applies eligibility rules, auto-approves clean cases, and routes the edge cases to an underwriter or adjuster.
Employee onboarding
Orchestrate onboarding end to end: provision accounts, assign equipment, schedule orientation, and notify every team — with approval gates exactly where they matter.
Automate your most complex processes.
Design, test, and run AI-native workflows on your own infrastructure. Book a walkthrough tailored to your use case.