Agent Bankz

Agent Bankz is an open-source, approval-gated AI revenue agent platform. It ships a local CLI, web dashboard, skills hub, connector registry, wallet command center, durable workflow design, and safety evals for building legitimate revenue workflows.

Agent Bankz is built for revenue creation, not reckless automation. It can research opportunities, package offers, draft sales assets, prepare listings, design launch experiments, and queue money or outreach actions for approval. It does not bypass credentials, send spam, execute unapproved trades, move money without consent, or claim regulated financial powers it does not have.

What ships in V1

• bankz CLI with init, up, chat, run, skills, connectors, models, model-runtime, security, tiers, subagents, benchmarks, updates, brand, whitepaper, wallet, eval, and temporal.
• Web dashboard with Mission Control, Revenue Lab, Skills Hub, Connector Center, Model Switcher, Wallet Command Center, Security Core, Stripe Paywall, Sub-Agent Runtime, Benchmark Cards, Browser Automation, Approval Queue, and Audit Log.
• Typed packages for agent policy, model runtime, tool routing, hosted-security scaffolding, subscription entitlements, sub-agent missions, benchmark cards, connector manifests, skills, wallet boundaries, and evals.
• 20 revenue-safe skills in skills/*/SKILL.md.
• Connector manifests for GitHub, Gmail, Google Workspace, Notion, Linear, Stripe, Plaid, HubSpot, Canva, Figma, Temporal, Cloudflare, Vercel, Netlify, Render, Neon, Hugging Face, Semrush, Alpaca, Binance public data, browser automation, IBM Quantum, Amazon Braket, and Azure Quantum.
• GitHub-ready scaffolding, Docker Compose, install scripts, CI, docs, and security policy.

Quick Start

npm install
npm run bankz -- init
npm run dev:web

Open http://localhost:5173.

Windows users can run:

powershell -ExecutionPolicy Bypass -File scripts/install.ps1

macOS/Linux users can run:

chmod +x scripts/install.sh
./scripts/install.sh

CLI Examples

npm run bankz -- skills
npm run bankz -- connectors
npm run bankz -- models
npm run bankz -- model-runtime
npm run bankz -- security
npm run bankz -- tiers
npm run bankz -- subagents
npm run bankz -- benchmarks
npm run bankz -- setup
npm run bankz -- setup-init
npm run bankz -- setup-stripe
npm run bankz -- wallet
npm run bankz -- eval
npm run bankz -- chat "Create a $49 digital product launch plan for independent consultants" --model openai:gpt-5.2

Verification

The default build path avoids native bundler workers, so it runs cleanly in restricted environments.

npm test
npm run build
npm run verify:runtime
npm run verify:secrets
npm run verify:web

Publish To GitHub

The repo is prepared for a main branch push. After GitHub CLI authentication, run:

scripts/publish-github.sh

Windows:

powershell -ExecutionPolicy Bypass -File scripts/publish-github.ps1

See docs/github-publishing.md.

Safety Model

Agent Bankz uses approval-gated autonomy:

• Research and local planning can run directly.
• Browser actions, outbound messages, payments, bank connections, crypto transfers, payouts, and trading actions require approval.
• Gmail is draft-first.
• Alpaca trading is paper-first.
• Stripe and Plaid start in test/sandbox mode.
• Quantum providers require real IBM Quantum, Amazon Braket, or Azure Quantum credentials.
• No security badge is displayed until a real external audit is complete.

Monetization

The open-source repo can be monetized through Stripe-backed hosted dashboards, premium connector packs, Pro skills, team audit controls, managed deployment, and a skills marketplace. See docs/monetization.md.

Production Setup

Run npm run bankz -- setup-init to create a private local .env, then npm run bankz -- setup to see model, Stripe, auth, vault, workflow, hosting, connector, and external-audit readiness. See docs/production-setup.md.

Research Grounding

The product direction is grounded in current agent, MCP, financial, and infrastructure patterns rather than fictional universal access. See docs/research/2050-agent-forecast.md.

License

Apache-2.0. See LICENSE.