GitHub - Tem-Degu/streetai-aaas: Agent As A Service (AaaS)

Turn what you know into a running business. No code required.

AaaS is an open protocol and toolkit for building AI agents that provide real services to real people through conversation. You don't write code. You don't design a UI. You describe what the agent should do, drop in your data, and connect it to a platform. The agent takes it from there.

• Describe the service by writing a skill document, or just tell the agent what you want and let it write one for you.
• Build the database by dropping JSON files into a folder, or by chatting with the agent and letting it organize the data itself. No schemas, no migrations, no code.
• Connect to users on Telegram, Discord, Slack, WhatsApp, your own website, or a social platform like Truuze. The agent handles every conversation, tracks every transaction, and grows its own knowledge over time.

The result is an agent that runs a service business on your behalf: it talks to customers, looks up data, proposes services, collects payments, delivers results, and remembers what it learned for next time.

Traditional SaaS:  Developer writes code  ->  deploys app    ->  users interact with UI
AaaS:              You share knowledge     ->  agent runs it  ->  users interact through chat

How It Works

An AaaS agent is built on seven pillars:

When a user messages your agent, it follows a structured lifecycle:

Explore  ->  Propose Service  ->  Create Transaction  ->  Deliver  ->  Complete

1. Explore: Understand what the user wants, check the data, assess feasibility
2. Propose Service: Make a plan, calculate cost, get user approval
3. Create Transaction: Register the job, start tracking
4. Deliver Service: Do the work, send the result
5. Complete Transaction: Confirm satisfaction, release payment

AaaS in Action

Maya lives in New York and knows the city's dating scene inside out — which neighborhoods click, which restaurants spark a real conversation, how to read between the lines of a profile. She doesn't know how to code, but she wants to turn that knowledge into a service.

1. She installs AaaS and creates an agent workspace
2. She writes the matchmaking skill: what the agent does, New York dating knowledge, pricing, boundaries
3. She seeds the database with initial profiles and venue data
4. She connects the agent to her preferred platforms

The agent is now live. When James messages asking for help finding a date, the agent explores his preferences, proposes a service tier, collects payment, delivers curated matches with compatibility notes, and logs the completed transaction. Maya earns money while the agent does the work.

Every interaction makes the service better: more profiles in the database means better matches, which means more satisfied users, which means more word of mouth.

Built-in Capabilities

Two capabilities ship with every agent and turn on as soon as you configure them.

Owner notifications

When something needs human judgment (a dispute, an unusual request, an extension that keeps failing) the agent reaches you on the channels you configure: Telegram, WhatsApp, or Email. Replies route back into the original customer conversation in admin mode, so a quick "approve refund" or "tell them no" on Telegram becomes the agent's next action with the customer.

# Configure channels in the dashboard's Notifications tab,
# or edit .aaas/notifications.json directly.

The agent decides when to reach out. Routine successes never trigger an alert.

Payments (Stripe)

Connect your own Stripe account in the dashboard's Payments tab and the agent gets six tools for taking and verifying payments: create_payment_request, get_payment_status, list_pending_payments, cancel_payment_request, refund_payment, list_payments. Three hard rules ship with the prompt so the agent cannot drift:

1. Never confirm a payment as received without verifying with Stripe in the same turn.
2. Only IDs returned by create_payment_request are real.
3. Refunds and discounts are owner-gated. The agent escalates via notify_owner rather than acting on a customer's word.

Money flows directly to your Stripe account. AaaS never holds funds. Test mode is the default until you switch to a sk_live_ key.

Quick Start

Install

npm install -g @streetai/aaas

Requires Node.js 18 or later.

Create an agent

Syntax:

aaas init <directory> [name] [description] [--type service|social]

• <directory> — folder name where the workspace will be created
• [name] — display name shown to users (optional, can be edited later)
• [description] — one-line summary (optional, can be edited later)
• --type — service (default, follows the transaction protocol) or social (creates content and engages in conversations)

Examples:

# Service agent — replace "my-agent" with the folder name you want
aaas init my-agent "Lyon Travel Guide" "Helps tourists explore Lyon, France"

# Social agent
aaas init my-bot "Aria" --type social

This creates a workspace with the full AaaS structure: skill template, soul file, data directory, extensions registry, and configuration.

Open the dashboard

The dashboard opens with a Setup Guide that walks you through configuring your LLM provider, adding data, writing your service definition, and deploying. You can also configure everything from the CLI:

# Configure LLM provider
aaas config --provider anthropic --model claude-sonnet-4-6 --key sk-ant-...

# Edit the skill file
aaas skill edit

Supported providers: Anthropic, OpenAI, Google, Ollama, OpenRouter, DeepSeek, Azure.

Connect to a platform

# HTTP API (simplest, includes embeddable chat widget)
aaas connect http --port 3300

# Telegram
aaas connect telegram --token YOUR_BOT_TOKEN

# Discord
aaas connect discord --token YOUR_BOT_TOKEN

# Slack
aaas connect slack --bot-token xoxb-... --app-token xapp-...

# WhatsApp (via WhatsApp Business Cloud API)
aaas connect whatsapp --access-token YOUR_ACCESS_TOKEN --phone-number-id YOUR_PHONE_NUMBER_ID --verify-token YOUR_VERIFY_TOKEN

# Truuze (social platform with native agent accounts)
# Three ways to connect:
#   1. With a provisioning SKILL.md downloaded from your Truuze account:
aaas connect truuze --skill ~/Downloads/SKILL.md
#   2. With a provisioning token directly:
aaas connect truuze --token YOUR_PROVISIONING_TOKEN
#   3. With an existing agent API key:
aaas connect truuze --key trz_agent_xxx

# OpenClaw (run inside an OpenClaw workspace)
aaas connect openclaw --id YOUR_AGENT_ID

# Public Relay (no public server required — also routes WhatsApp + chat widget)
aaas connect relay

Start serving

Your agent is now live on all connected platforms. Users message it, and it follows the AaaS protocol to serve them.

To start only a subset of platforms, pass their names:

aaas run telegram
aaas run telegram discord

Add --daemon to run in the background:

aaas run telegram --daemon

If a daemon is already running when you use --daemon with a platform filter, you'll be prompted to stop it and start a fresh daemon with the listed platforms.

Check everything is working

Verifies node version, credentials, LLM reachability, connections, workspace structure, and more.

Open the dashboard

Opens the web dashboard for the specified agent. You can also run aaas dashboard from inside a workspace directory, or with no arguments to open the hub dashboard showing all your agents.

Chat Widget

The fastest way to put your agent on any website is the public Relay. Connect the relay first to get your unique slug, then drop one script tag into your HTML before the closing </body>:

# Register your agent with streetai.org and get a public slug
aaas connect relay

<script
  src="https://streetai.org/a/YOUR_SLUG/widget.js"
  data-agent="https://streetai.org/a/YOUR_SLUG"
  data-title="Ask me anything about Lyon"
  data-color="#2563eb"
  data-position="right"
  data-greeting="Bonjour! How can I help you explore Lyon?"
></script>

Visitors chat through streetai.org, which forwards messages over WebSocket to your locally-running agent — no public IP, no port forwarding, no build step. The widget renders a floating chat button, supports file attachments (images, audio, video, PDFs), and persists conversation history per visitor.

Widget options:

CLI Commands

Workspace

Content

Configuration

Data

Transactions

Extensions

Platform Connections

Agent Lifecycle

Hub (Multi-Agent Management)

Extensions

Agents can call external APIs, other agents, human contacts, and local tools through extensions. The extension system supports multiple auth types (Bearer, custom header, query parameter, Basic auth), custom headers, and all HTTP methods.

Extensions also enable payment flows. When a service involves payments through an external provider like Stripe or PayPal, the agent creates a payment link via the extension, sends it to the user, and verifies the payment status when the user confirms.

{
  "name": "Stripe Payments",
  "type": "api",
  "endpoint": "https://api.stripe.com/v1",
  "auth": {
    "type": "bearer",
    "apiKey": "sk_live_..."
  },
  "capabilities": ["create_payment_link", "verify_payment"],
  "cost_model": "per_request"
}

See docs/extensions.md for the full spec.

Dashboard

The web dashboard gives you a complete view of your running agent:

• Overview: Revenue, active/completed transactions, connected platforms, memory stats
• Skill & Soul: View and edit the agent's knowledge base and personality
• Data: Browse and manage the service database
• Transactions: Full history with detail views, filtering, and revenue breakdowns
• Extensions: Register, configure, and test extensions with all auth types
• Memory: View the agent's stored facts
• Connections: See which platforms are connected
• Chat: Test conversations with the agent directly
• Guide: Setup instructions for each connector
• Deploy: Deployment options and API endpoint reference

Project Structure

aaas/
├── src/
│   ├── cli/                  # CLI commands (init, config, run, connect, etc.)
│   ├── connectors/           # Platform connectors (HTTP, Telegram, Discord, etc.)
│   ├── engine/               # Core agent engine, prompts, and tool definitions
│   ├── server/               # Dashboard API server
│   ├── widget/               # Embeddable chat widget
│   ├── auth/                 # Authentication utilities
│   └── utils/                # Shared helpers
├── dashboard/                # React web dashboard (Vite + React)
│   ├── src/pages/            # Dashboard pages
│   └── dist/                 # Pre-built dashboard (shipped with npm package)
├── templates/
│   └── workspace/            # Scaffold used by `aaas init` (SOUL.md, data/, extensions/, etc.)
├── docs/                     # Protocol documentation
│   ├── protocol.md           # Full protocol specification
│   ├── extensions.md         # Extension protocol and payment flows
│   ├── skill-reference.md    # How to write a skill
│   ├── transactions.md       # Transaction lifecycle
│   └── ...
├── bin/                      # Helper scripts (scaffold.sh)
└── examples/                 # Example agent workspaces

Platform Support

AaaS ships with connectors for six platforms, a general-purpose HTTP API, and a relay for serverless deployments:

You can connect to multiple platforms at the same time. Run aaas run and the agent serves on all of them.

Relay (streetai.org)

If you don't have a public server, use the relay. It proxies WhatsApp webhooks and chat widget traffic through streetai.org to your locally-running agent via WebSocket.

# Connect WhatsApp credentials (stored locally, never sent to relay)
aaas connect whatsapp --access-token TOKEN --phone-number-id ID --verify-token SECRET

# Register with the relay
aaas connect relay

# Start — agent connects outbound to streetai.org, no public IP needed
aaas run

The relay gives you public URLs for your chat widget and WhatsApp webhook. Embed the widget on any website, paste the webhook URL into Meta's dashboard, and you're live. The chat widget supports file attachments — files are uploaded to the relay and forwarded to your agent.

Earn on Truuze

Truuze is a social platform built for AI agents to deliver paid services. When you connect there, your agent gets a public profile, a chat inbox, and an escrow-protected way to take payment.

The flow is short:

1. Build with AaaS. Run aaas dashboard to set up the service in chat, drop reference data into the workspace, and add any extensions the agent needs (other agents or external API calls).
2. Generate a skill on Truuze. Open app.truuze.com, create an account, go to the AI Agents tab, and click Add New to download a SKILL.md.
3. Connect and start. From the Deploy tab in your dashboard, connect using that SKILL.md (or paste an existing agent API key), then click Start.

Customers fund escrow when they accept an offer, and funds release once the delivery is approved. If a dispute is raised, the agent has 48 hours to resolve it directly with the customer; after that window, a Truuze admin steps in. Truuze handles the service and payment lifecycle so you can focus on building a great agent.

Full walkthrough: streetai.org/docs/truuze.html

Contributing

This is an early-stage project. Contributions are welcome:

• Protocol improvements: Open an issue to discuss changes to the spec
• New examples: Submit example agents for different service domains
• Connectors: Build support for new platforms
• Documentation: Improve guides, fix errors, add translations

License

Apache-2.0. See LICENSE.