____ _ ___ __ _ _
| __ ) __ _ _ __ | | __ / _ \ / _| / \ __ _ ___ _ __ | |_ ___
| _ \ / _` | '_ \| |/ / | | | | |_ / _ \ / _` |/ _ \ '_ \| __/ __|
| |_) | (_| | | | | < | |_| | _| / ___ \ (_| | __/ | | | |_\__ \
|____/ \__,_|_| |_|_|\_\ \___/|_| /_/ \_\__, |\___|_| |_|\__|___/
|___/
Your AI-Powered C-Suite. One Human CEO, Unlimited Scale.
The world's first platform where AI agents operate as autonomous department heads --
each with real budgets, real spending authority, and real accountability on-chain.
┌───────────────────────────────────────────────────────────────────┐
│ BANK OF AGENTS │
│ │
│ ┌──────────────┐ ┌──────────────┐ ┌──────────────┐ │
│ │ Vite SPA │ │ Hono API │ │ OpenClaw │ │
│ │ React 19 │◄─►│ Server │◄─►│ Gateway │ │
│ └──────┬───────┘ └──────┬───────┘ └──────┬───────┘ │
│ │ │ │ │
│ ┌──────┴───────┐ ┌──────┴───────┐ ┌──────┴───────┐ │
│ │ RainbowKit │ │ Supabase │ │ Claude │ │
│ │ Monad │ │ Drizzle │ │ Sonnet/Haiku│ │
│ └──────┬───────┘ └──────────────┘ └──────┬───────┘ │
│ │ │ │
│ ┌──────┴───────┐ ┌──────┴───────┐ │
│ │ x402 │ │ MCP │ │
│ │ USDC │ │ Servers │ │
│ └──────────────┘ └──────────────┘ │
│ │
│ Governance: 5-Layer Safety Engine (Deterministic, No LLM) │
│ Settlement: Monad -- 10,000 TPS, 0.4s Finality, Native USDC │
└───────────────────────────────────────────────────────────────────┘
- AI tools stop at recommendations. They tell you what to do -- "run a Google Ads campaign," "deploy on Railway" -- and hand execution back to you. The human remains the bottleneck for every operational decision.
- Budget management is binary. Existing agent solutions use hard spending caps with zero awareness of ROI, velocity, or cross-department optimization. A $500 spend with 10x return is treated the same as a $500 loss.
- Solopreneurs need 50-person capacity. The founder-first economy is here, but one person still cannot run marketing, engineering, HR, and finance simultaneously.
- Agents that act, not advise. When the Marketing agent determines a Runway subscription will improve ad performance, it signs up, creates the video, pays with USDC on Monad, and reports the outcome.
- ROI-aware autonomous spending. Agents track return on every transaction, learn from spending patterns, and adjust future decisions. The Finance agent monitors all departments in real-time.
- Blockchain-native settlement. Every payment settles on Monad via the x402 protocol -- 10,000 TPS, 0.4-second finality, per-agent wallets, complete fund separation.
No existing product combines all three: departmental agent orchestration (agents as a functional org chart), ROI-aware autonomous budgets (spending intelligence beyond hard caps), and blockchain-native payments (real settlement on a high-performance chain). Bank of Agents sits at the intersection of all three.
Real-time overview of all agent activity, budgets, and performance. Agent summary cards with live status indicators, budget utilization bars, ROI percentages, and a unified activity feed streaming across all departments.
Assign agents to projects, set goals, allocate budgets per initiative, and track progress. Project-level financial controls with task assignment and cross-agent collaboration.
Stream an agent's reasoning in real-time. Thought logs, decision trails, transaction history, and an intervention bar for direct CEO commands. See exactly why an agent chose vendor A over vendor B.
17 agents across 4 categories, each with a personality (SOUL.md), specialties, and tracked ROI. Deploy immediately or create custom agents. Browse, evaluate, and hire in seconds.
Every payment includes the agent's reasoning, ROI projection, governance check results, and on-chain settlement details. Filter by agent, status, category, or date range. Export for accounting.
Spending timelines, burn rate charts, ROI comparisons, budget runway gauges, agent performance radar, and spending donuts. Powered by Recharts and TradingView Lightweight Charts.
Configure LLM API keys, select models (Sonnet/Haiku), set governance thresholds, manage deployment modes, toggle light/dark theme, and control per-agent operating modes.
17 agents with distinct personalities defined in SOUL.md files. 6 shared skill modules defined in SKILL.md files. Each agent has its own wallet, budget, and domain expertise.
| Agent | Code Name | Domain | Color |
|---|---|---|---|
| Marketing | The Strategist | Ad optimization, ROI analysis, campaign management, creative tools | #8C7B6B |
| Dev | The Builder | Cloud infrastructure, CI/CD, deployment, cost optimization | #7B8C6B |
| HR | The Recruiter | Talent acquisition, job boards, candidate screening, interview scheduling | #6B7B8C |
| Finance | The Guardian | Budget oversight, anomaly detection, spending analytics, runway projection | #B08D57 |
| Researcher | The Sentinel | Vendor safety verification, domain analysis, fraud detection, trust scoring | #8C6B6B |
| Agent | Code Name | Domain |
|---|---|---|
| Personal Finance | The Treasurer | Bill tracking, subscription cancellation, savings automation |
| Wellness | The Coach | Gym memberships, health appointments, supplement management |
| Learning | The Scholar | Online courses, book curation, language learning, certifications |
| Travel | The Navigator | Flight deals, hotel comparison, experience booking, price monitoring |
| Agent | Code Name | Domain |
|---|---|---|
| Creator | The Muse | Creative tools, AI generation credits, stock assets, publishing |
| Freelancer | The Operator | Invoicing, project management, expense tracking, tax prep |
| Smart Home | The Automator | Security subscriptions, streaming bundles, cloud storage, IoT |
| Investor | The Oracle | Market data feeds, research platforms, financial newsletters |
| Agent | Code Name | Domain |
|---|---|---|
| Social | The Connector | Restaurant reservations, event tickets, gift purchases |
| Parent | The Caretaker | Children's activities, school fees, childcare scheduling |
| Pet Care | The Companion | Food deliveries, vet appointments, grooming, pet insurance |
| Gamer | The Grinder | Game subscriptions, sale tracking, microtransaction control |
Every agent can access these modular capabilities:
| Skill | Purpose |
|---|---|
| x402 Payment | Sign and settle USDC transactions on Monad |
| Web Research | Search, browse, and extract data from the web |
| Budget Management | Check balances, request transfers, track spending |
| Service Evaluation | Compare vendors, calculate ROI, score alternatives |
| Collaboration | Request help from other agents, delegate tasks |
| Safety Verification | Multi-stage vendor verification before payments (domain, SSL, WHOIS, pricing, trust scoring) |
| Layer | Technology | Purpose |
|---|---|---|
| Frontend | Vite 8 + React 19 + TypeScript 5.9 | Single-page application |
| Routing | React Router v7 | Client-side SPA routing with lazy loading |
| Styling | Tailwind CSS v4.2 | Design system with @theme tokens |
| Animation | Motion v12 + Rive + tailwindcss-motion | Micro-interactions and agent animations |
| State | TanStack Query v5 + Zustand v5 | Server state + client state |
| Forms | React Hook Form + Zod v4 | Validated budget and config forms |
| Visualization | Recharts + TradingView Lightweight Charts | Financial charts and analytics |
| 3D | React Three Fiber + Spline | Visual effects |
| Web3 | wagmi v3 + viem v2 + RainbowKit v2 | Monad wallet connection |
| Payments | x402 protocol (core + evm + fetch) | Autonomous USDC payments |
| Blockchain | Monad (Chain ID 143 / 10143) | 10K TPS, 0.4s blocks, EVM-compatible |
| Backend | Hono 4 + Node.js | API server with Hono RPC |
| Database | Supabase (PostgreSQL) + Drizzle ORM | Relational data and migrations |
| Cache | Redis + BullMQ | Event bus, job queues, agent messaging |
| AI | Anthropic Claude (Sonnet 4 / Haiku) | Agent reasoning and decision-making |
| Observability | Langfuse | LLM tracing, token costs, latency |
| Orchestration | OpenClaw | Multi-agent gateway and routing |
| Tools | Model Context Protocol (MCP) | Agent tool servers (payments, budgets, browse) |
| Real-time | SSE + WebSocket | Live agent streams and dashboard updates |
Agent LLM decides to pay
│
v
┌─────────────────────────────────────────────────────┐
│ SAFETY VERIFICATION LAYER │
│ │
│ Stage 1: Whitelist/Blacklist (< 50ms) │
│ Stage 2: 6 Parallel Checks (domain, SSL, WHOIS, │
│ pricing, trust, URL validation) │
│ Stage 3: Researcher Agent (LLM analysis) │
│ Stage 4: Human Escalation (if borderline) │
└─────────────────────────┬───────────────────────────┘
│
PASS / BLOCK
│
v
┌─────────────────────────────────────────────────────┐
│ 5-LAYER GOVERNANCE ENGINE │
│ │
│ Layer 1: Agent Constraints │
│ Mandate-scoped categories, per-tx limits │
│ │ │
│ Layer 2: Budget Check │
│ Balance verification, overage detection │
│ │ │
│ Layer 3: Velocity Check │
│ Spending rate analysis, duplicate detection │
│ │ │
│ Layer 4: Cross-Agent Review │
│ Finance agent reviews large purchases │
│ │ │
│ Layer 5: Global Safeguards │
│ System caps, circuit breakers, kill switch │
└─────────────────────────┬───────────────────────────┘
│
APPROVED / REJECTED
│
v
EIP-712 Typed Data Signature
(Agent wallet or CEO co-sign)
│
v
x402 Facilitator (molandak.org)
│
v
Monad Settlement (0.4s finality)
Native USDC, on-chain receipt
│
v
SSE push to dashboard (real-time)
All governance checks are deterministic TypeScript -- no LLM reasoning touches budget math or payment execution. Each agent has its own wallet with complete fund separation. Co-pilot mode transactions above the configured threshold require the human CEO's wallet signature.
Every agent supports two modes, configurable by the CEO:
| Mode | Behavior | Best For |
|---|---|---|
| Autopilot | Fully autonomous within allocated budget. No human approval required for individual transactions. | Trusted agents with proven ROI. Day-to-day operational spending. |
| Co-pilot | Autonomous below a configurable threshold. Requires wallet signature above it. | New agents, large purchases, domains requiring oversight. |
Bank of Agents supports three deployment modes that can be mixed per agent:
All agents on one machine. Docker Compose or bare Node.js processes behind a single OpenClaw Gateway. Zero recurring cost. In-process event transport with ~0.01ms latency.
Each agent in its own container on Railway or Fly.io. Always-on, independently scalable. Redis-based event transport. Best for production workloads.
Privacy-sensitive agents run locally while business-critical agents run in the cloud. Coordinated via Upstash Redis with secure tunneling. Mix and match per agent.
Local Pack Cloud Fleet Edge Hybrid
┌────────────────┐ ┌────────────────┐ ┌────────────────┐
│ All agents │ │ Agent per │ │ Local agents │
│ One machine │ │ container │ │ + Cloud agents│
│ Docker Compose│ │ Railway/Fly │ │ Redis bridge │
│ ~$0/month │ │ ~$25/month │ │ Best of both │
└────────────────┘ └────────────────┘ └────────────────┘
Every agent, regardless of deployment mode, settles transactions on the same Monad chain using USDC via x402. Deployment mode affects latency, availability, and privacy -- never payment capability.
Agents are orchestrated through the OpenClaw gateway:
- SOUL.md -- Each agent's personality, expertise, constraints, and spending philosophy
- SKILL.md -- Shared capabilities (payments, research, budgets, evaluation, collaboration)
- MCP Servers -- Tool servers for payments, budget queries, and web browsing
- Multi-agent routing -- Gateway routes messages between agents with location transparency
- Agent mesh -- Peer-to-peer communication with request/response, broadcast, and capability-based routing
The mesh abstracts agent location entirely. An agent sends a message by ID; the router resolves transport automatically -- from in-process function calls to cross-region Redis pub/sub.
bank-of-agents/
├── packages/
│ ├── web/ # Vite SPA -- React 19, TypeScript, Tailwind v4
│ │ ├── src/
│ │ │ ├── pages/ # 8 pages (Dashboard, Workspace, Agents, Marketplace, ...)
│ │ │ ├── components/ # 80+ React components organized by domain
│ │ │ │ ├── dashboard/ # Stats, budget bars, agent cards, activity feed
│ │ │ │ ├── agents/ # Live view, thought log, decision trail, screen
│ │ │ │ ├── marketplace/ # Agent cards, hire modal, create agent, deploy
│ │ │ │ ├── transactions/ # Table, filters, export, status badges
│ │ │ │ ├── analytics/ # Charts, radar, donut, burn rate, ROI
│ │ │ │ ├── workspace/ # Projects, tasks, budgets, agent assignment
│ │ │ │ ├── deployment/ # Fleet map, mode selector, cost calculator
│ │ │ │ ├── wallet/ # RainbowKit connect button
│ │ │ │ └── ui/ # Shared primitives (Card, Badge, Button, Skeleton)
│ │ │ ├── hooks/ # TanStack Query hooks, SSE, simulation, chat
│ │ │ ├── stores/ # Zustand stores (agents, LLM config)
│ │ │ ├── data/ # Mock data for all domains
│ │ │ ├── lib/ # Utilities, wagmi config, API client, formatting
│ │ │ └── providers/ # Web3Provider (wagmi + RainbowKit + Monad)
│ │ └── package.json
│ │
│ └── api/ # Hono API Server -- TypeScript, Node.js
│ ├── src/
│ │ ├── routes/ # 16 route modules (agents, budgets, payments, ...)
│ │ ├── agents/ # Runtime, personalities, LLM client, tool registry
│ │ ├── governance/ # 5-layer engine (constraints, budget, velocity, cross, global)
│ │ ├── payments/ # x402 client, Monad config, per-agent wallets
│ │ ├── simulation/ # Behavior engine with per-agent simulators
│ │ ├── realtime/ # SSE event bus, WebSocket handlers
│ │ ├── mesh/ # Agent-to-agent communication (router, transports)
│ │ ├── a2a/ # Agent cards, task executor, registry
│ │ ├── openclaw/ # Gateway bridge, client, fallback
│ │ ├── mcp/ # MCP servers (payment, budget, browse)
│ │ ├── safety/ # Vendor verification pipeline with 6 parallel checks
│ │ ├── browser/ # Playwright-based agent web browsing
│ │ ├── discovery/ # Cross-device agent registry and coordination
│ │ ├── gateway/ # Proxy and SSE aggregation for multi-container mode
│ │ ├── logging/ # Activity, audit, and transaction loggers
│ │ ├── analytics/ # Metrics computation engine
│ │ ├── budgets/ # Centralized budget manager
│ │ ├── status/ # Real-time agent status tracking
│ │ ├── thoughts/ # Agent reasoning chain storage
│ │ ├── deployment/ # Manager, migration, health checks, cost calculator
│ │ ├── middleware/ # Rate limiting, validation, governance
│ │ └── db/ # Drizzle schema, relations, seed data
│ └── package.json
│
├── agents/ # 17 SOUL.md + 6 SKILL.md personality files
│ ├── marketing/SOUL.md # "The Strategist"
│ ├── dev/SOUL.md # "The Builder"
│ ├── hr/SOUL.md # "The Recruiter"
│ ├── finance/SOUL.md # "The Guardian"
│ ├── personal-finance/SOUL.md # "The Treasurer"
│ ├── wellness/SOUL.md # "The Coach"
│ ├── learning/SOUL.md # "The Scholar"
│ ├── travel/SOUL.md # "The Navigator"
│ ├── creator/SOUL.md # "The Muse"
│ ├── freelancer/SOUL.md # "The Operator"
│ ├── smart-home/SOUL.md # "The Automator"
│ ├── investor/SOUL.md # "The Oracle"
│ ├── social/SOUL.md # "The Connector"
│ ├── parent/SOUL.md # "The Caretaker"
│ ├── pet-care/SOUL.md # "The Companion"
│ ├── gamer/SOUL.md # "The Grinder"
│ ├── researcher/SOUL.md # "The Sentinel"
│ └── skills/
│ ├── x402-payment/SKILL.md
│ ├── web-research/SKILL.md
│ ├── budget-management/SKILL.md
│ ├── service-evaluation/SKILL.md
│ ├── collaboration/SKILL.md
│ └── safety-verification/SKILL.md
│
├── docs/ # 12 design documents
│ ├── PROJECT.md # Vision, competitive analysis, target audience
│ ├── ARCHITECTURE.md # System design, data flow, security model
│ ├── TECH_STACK.md # Technology decisions with rationale
│ ├── REQUIREMENTS.md # Functional and non-functional requirements
│ ├── DEPLOYMENT_ARCHITECTURE.md # Local Pack / Cloud Fleet / Edge Hybrid
│ ├── DEPLOYMENT_CONFIG.md # Configuration profiles and CLI
│ ├── AGENT_MESH.md # Agent-to-agent communication protocol
│ ├── UI_UX_DESIGN.md # Design system, component specs
│ ├── TYPOGRAPHY_RESEARCH.md # Font selection rationale
│ ├── SAFETY_VERIFICATION.md # Vendor verification architecture and threat model
│ ├── BROWSER_AUTOMATION.md # Playwright browser automation design
│ └── BACKEND_IMPROVEMENT_PLAN.md # Backend roadmap and priorities
│
├── DEPLOYMENT.md # Production deployment guide
├── docker-compose.yml # Docker Compose for local development
└── package.json # pnpm monorepo root
By the numbers: ~130 frontend source files, ~170 backend source files, 17 agent personalities, 6 shared skills, 12 design documents.
- Node.js >= 20.0.0
- pnpm -- install via
corepack enable
git clone https://github.com/pancodo/BankOfAgents.git
cd bank-of-agents
pnpm install# Start the frontend (http://localhost:5173)
pnpm dev
# Start the API server (http://localhost:3001)
pnpm dev:api
# Start both in parallel
pnpm dev:allpnpm build # Frontend production build (tsc + vite)
pnpm build:api # API production build (tsup)pnpm --filter api db:generate # Generate Drizzle migrations
pnpm --filter api db:migrate # Run migrations
pnpm --filter api db:seed # Seed sample data
pnpm --filter api db:studio # Open Drizzle Studiopnpm docker:up # Start full stack (web, API, Postgres, Redis)
pnpm docker:down # Stop all services
pnpm docker:logs # Tail logs| Variable | Required | Description |
|---|---|---|
DATABASE_URL |
Yes | Supabase PostgreSQL connection string |
REDIS_URL |
Yes | Upstash Redis URL |
ANTHROPIC_API_KEY |
Yes | Anthropic API key for Claude |
CORS_ORIGIN |
Yes | Allowed frontend origins (comma-separated) |
PORT |
No | Server port (default: 3001) |
NODE_ENV |
No | development or production |
MONAD_RPC_URL |
No | Monad RPC endpoint (default: testnet) |
PRIVATE_KEY |
No | Wallet private key for x402 payments |
X402_FACILITATOR_URL |
No | x402 facilitator service URL |
OPENCLAW_ENABLED |
No | Enable OpenClaw gateway integration |
SIMULATION_AUTOSTART |
No | Start simulation engine on boot |
BROWSER_ENABLED |
No | Enable Playwright browser automation |
BROWSER_HEADLESS |
No | Run browser in headless mode (default: true) |
DISCOVERY_ENABLED |
No | Enable cross-device agent discovery |
DEVICE_ID |
No | Unique device identifier for discovery |
DEVICE_MODE |
No | Operating mode: all, single, or gateway |
AGENT_ROLE |
No | Agent role for single-container mode |
| Variable | Required | Description |
|---|---|---|
VITE_API_URL |
Yes | Hono API server base URL |
VITE_WALLETCONNECT_PROJECT_ID |
Yes | WalletConnect Cloud project ID |
The "Private Bank" aesthetic draws from high-end financial interfaces -- warm, confident, and readable. Light and dark modes with a premium paper feel.
| Element | Value |
|---|---|
| Canvas | #FAFAF7 -- warm cream, not cold white |
| Surface | #FFFFFF -- cards on cream background |
| Accent | #B08D57 -- champagne gold |
| Text | #1A1612 -- warm dark, not pure black |
| Headings | Instrument Serif (italic) |
| Body | Geist sans-serif |
| Mono | Geist Mono (tabular-nums for financial figures) |
Agent colors are muted earth tones designed to feel professional: #8C7B6B marketing, #7B8C6B dev, #6B7B8C HR, #B08D57 finance.
Animations: Motion v12 with 60ms stagger delays and 300ms page transitions. Cards use warm shadows. Dark mode inverts the palette while preserving the premium feel.
- Frontend MVP -- Dashboard, Workspace, Agents, Marketplace, Transactions, Analytics, Settings
- Design system -- Private Bank aesthetic with light and dark mode
- Backend foundation -- Hono API server, Drizzle schema, 16 route modules
- Agent runtime -- Personalities, tool registry, LLM integration, simulation engine
- 5-layer governance engine -- Deterministic TypeScript, no LLM in safety checks
- x402 payment integration -- Client, per-agent wallets, Monad configuration
- 17 agent personalities -- SOUL.md files with full identity definitions
- 6 shared skill modules -- SKILL.md files for cross-agent capabilities
- Real-time infrastructure -- SSE event bus, WebSocket handlers
- Agent mesh -- Communication protocol with transport abstraction
- Deployment architecture -- Local Pack, Cloud Fleet, Edge Hybrid
- OpenClaw integration -- Gateway bridge, client, fallback routing
- MCP servers -- Payment, budget, and browse tool servers
- Docker Compose -- Full local development stack
- Safety verification pipeline -- 6 parallel checks, trust lists, researcher agent
- Browser automation -- Playwright-based agent web browsing with security sandbox
- Distributed discovery -- Cross-device agent registry with Redis coordination
- Gateway mode -- Multi-container deployment with SSE aggregation
- Analytics engine -- Overview metrics, burn rate, agent scorecards
- Logging system -- Activity, audit (SHA-256 chain), and transaction loggers
- Budget manager -- Centralized tracking with velocity alerts and runway projection
- Agent thought/decision stores -- Reasoning chains with ring buffers
- Frontend-backend integration via Hono RPC client
- Production deployment -- Vercel (frontend) + Railway (backend)
- Live x402 payments on Monad testnet
- Agent character illustrations (AI-generated)
- Community agent marketplace
- Mobile-responsive optimization
- Mainnet deployment
| Document | Description |
|---|---|
| PROJECT.md | Vision, competitive analysis, target audience, success metrics |
| ARCHITECTURE.md | System design, governance engine, data flow, security model |
| TECH_STACK.md | Every technology choice with rationale |
| REQUIREMENTS.md | Complete functional and non-functional requirements |
| DEPLOYMENT_ARCHITECTURE.md | Local Pack, Cloud Fleet, Edge Hybrid specifications |
| DEPLOYMENT_CONFIG.md | Configuration profiles, CLI, and YAML schemas |
| AGENT_MESH.md | Agent-to-agent communication protocol |
| UI_UX_DESIGN.md | Design system, component specifications |
| SAFETY_VERIFICATION.md | Vendor verification architecture and threat model |
| BROWSER_AUTOMATION.md | Playwright browser automation design |
| BACKEND_IMPROVEMENT_PLAN.md | Backend roadmap and priorities |
| DEPLOYMENT.md | Production deployment guide (Vercel + Railway + Supabase) |
| Monad High-performance EVM 10K TPS, 0.4s blocks |
x402 Payment protocol Native USDC settlement |
OpenClaw Agent orchestration Multi-agent gateway |
Anthropic Claude Sonnet / Haiku Agent reasoning |
| Supabase PostgreSQL database Auth and real-time |
Vite Build tool Sub-50ms HMR |
Hono API framework Type-safe RPC |
Tailwind CSS Utility-first CSS v4 design tokens |
Contributions are welcome. Please open an issue to discuss proposed changes before submitting a pull request.
- Fork the repository
- Create a feature branch (
git checkout -b feature/agent-name) - Make your changes
- Run
pnpm build && pnpm build:apito verify builds pass - Open a pull request with a clear description
Conventions: pnpm for package management. Double quotes in TSX, single quotes in config. PascalCase component files. Tailwind v4 with @theme tokens. Strict TypeScript throughout.
Bank of Agents -- because the future of business is not about hiring more people. It is about deploying smarter agents.