Approval workflows,defined as code.
ApprovalML is the open workflow definition language and runtime underneath Aptiwise. Steps, routing, forms, and SLAs compile down to portable YAML — reviewable in a pull request like any other config file — and it runs on your own infrastructure if you want it to.
The Problem
Approval logic shouldn't be a black box you can't audit or run yourself.
Workflow logic locked in a vendor's UI
Most approval tools store routing logic in a proprietary database you can't diff, review, or move — there's no way to see what changed or roll it back.
Compliance-sensitive data can't leave your network
Regulated teams need the approval engine running on their own infrastructure — not a hard dependency on someone else's cloud.
AI agents can act with no human checkpoint
Autonomous agents given tool access can create, overwrite, or delete — with no gate in front of the sensitive calls and no audit trail of what was requested.
How ApprovalML Works
The same engine that powers Aptiwise — on your terms
Workflows compile to data, not code — so they're portable, diffable, and safe to hand to a human or an AI agent.
Steps, routing, forms, and SLAs — all defined as data
Every workflow compiles to portable, version-controllable ApprovalML YAML. Review a routing change in a pull request the same way you'd review any other config — no proprietary format, no black-box UI required to understand what will happen.
- Human decision steps, notifications, and conditional routing
- Parallel approvals and child-workflow spawning for fan-out
- Asset registry with JSON record storage in PostgreSQL
- id: manager_approval
type: decision
sla: 24h
notify: email
A human checkpoint in front of any AI agent
An agent requests approval over MCP; a human approves or rejects by email, no login required. The agent's tool call only proceeds once a human signs off — routed through the identical engine as every other ApprovalML workflow.
- Email approve/reject, no login required
- The agent's tool call only executes after approval
- Full details in the AI Governance use case
Run the same engine that powers Aptiwise, on your own infrastructure
One Docker command stands up the full open-source approval engine and email approvals — no external cloud dependency required. Start self-hosted, and move to Aptiwise Cloud later without rewriting a single workflow.
The open-source runtime covers the full approval engine and email approvals. AI-powered workflow generation, the admin UI, and multi-tenant management are part of Aptiwise Cloud.
Wrap any MCP server — including ones you didn't build
One command spawns any existing stdio MCP server as a child process and re-exposes it with sensitive tools gated behind human approval — no changes to the upstream server, no per-server integration work.
Self-Host or Go Managed
One engine, two ways to run it
ApprovalML (Open Source)
- Full approval engine & email approvals
- REST API and MCP server access
- Deploys with Docker on your own infrastructure
Aptiwise Cloud
- Everything in ApprovalML, fully managed
- AI-powered workflow generation & admin UI
- Multi-tenant, multi-company management
ApprovalML
Run it yourself,
or let us run it for you.
Self-host the open-source runtime today, or start on Aptiwise Cloud and get the full AI-powered platform on top of the same engine.