Usage-based and hybrid pricing has made billing operationally heavy: many rates, usage dimensions, credits, commitments, and frequent pricing changes. The result is revenue leakage — under-monetized usage, margin compression, and failed-payment loss — that manual RevOps teams catch too slowly. This isn't a dashboard problem; it's a cashflow and margin-protection problem.

OpenRevOps is a self-hostable decision-and-execution layer that sits between usage telemetry and your billing system. It is built around a transparent, auditable agent loop:

ingest → detect risk/opportunity → recommend (with rationale + confidence)
 → enforce policy + human approval gate → execute via connector → audit

It is not a payment processor and not a replacement for your ERP or revenue-recognition stack. Stripe provides the billing primitives; OpenRevOps provides the decision intelligence, policy/guardrail orchestration, and auditable approvals over those primitives.

• Approval-safe by default. Irreversible actions require an explicit human approval gate. There is a hard kill-switch to require approval for everything.
• Transparent. Every action produces a human-readable decision trace and an append-only audit record.
• Reproducible. Core demo flows run end-to-end on synthetic data with a fixed seed — no private corpus required.
• Extensible. Connectors and policy rules are modular adapters.

open-revops/
 apps/
 web/ # Next.js operator UI (recommendations, approvals, audit timeline)
 api/ # Agent API + orchestration
 packages/
 agent-core/ # Planning, decision policy, execution graph
 policy-engine/ # Guardrails, approval gates, action thresholds
 connectors-stripe/ # Billing / checkout / dispute action adapters
 connectors-exa/ # Search / research context adapters
 connectors-storage/ # DB / event-store / audit-log adapters
 domain-pricing/ # Outcome-to-Invoice domain logic
 domain-renewals/ # Renewal-negotiation domain logic
 evals/ # Replay / evaluation harness for agent decisions
 demo-data/ # Synthetic accounts / usage events (deterministic)
 docs/ # Architecture, threat model, decision-trace spec, demo script

See docs/architecture.md and docs/threat-model.md.

Prerequisites: Node 22+ (.nvmrc), pnpm 10+ (via corepack enable), and a Stripe test-mode account. Use sandbox/test keys only.

git clone https://github.com/mvincentong/open-revops.git
cd open-revops
# Use the pinned Node version and package manager
nvm use # or: fnm use
corepack enable
# Configure environment (never commit your .env)
cp .env.example .env # then fill in TEST keys
# Install workspace dependencies
pnpm install

The standardized command surface every package exposes (see CONTRIBUTING):

pnpm lint # lint + format check
pnpm typecheck # type checks
pnpm test # unit tests
pnpm build # build all packages
pnpm demo:seed # seed deterministic synthetic data (once implemented)
pnpm demo:reset # reset demo to a clean state
pnpm dev # run the operator UI + API locally

While the repo is in the scaffold phase these scripts are placeholders that exit successfully; they are filled in as packages land.

Security is a first-class concern because this project touches billing actions:

• No secrets in the repo. Secrets are read from the environment only. .env is gitignored; .env.example documents required variables. Secret scanning runs in CI.
• Webhook verification. Stripe webhook signatures are verified before processing.
• Approval gate. Irreversible actions are gated behind explicit human approval.
• Audit trail. Append-only decision/action records with redaction of sensitive fields.

Found a vulnerability? Do not open a public issue. See SECURITY.md for private disclosure via GitHub Security Advisories.

We welcome contributions! Please read:

• CONTRIBUTING.md — local setup, branch/commit style, PR checklist
• AGENTS.md — rules for AI coding agents and humans (scope, sources, prohibited copying, security/data handling)
• CODE_OF_CONDUCT.md
• docs/ — architecture and specs

This repo is pre-wired for Claude Code with the Superpowers and ECC plugins. See .claude/README.md for how that works and the trust prompt you will see on first open.

Apache-2.0 © 2026 OpenRevOps Contributors. See NOTICE for third-party attributions.