Skip to content

PSAgents - End-User System Usage Documentation

Welcome to PSAgents, the central multi-tenant Agentic OS designed to configure, monitor, and scale your autonomous AI agent fleet. This guide details the step-by-step workflow for business users to operate the dashboard.

1. Getting Started: Signup & Workspace Initialization

  1. Accessing the Portal: Navigate to https://agents.pluggedspace.org/login.
  2. Registration:
  3. Toggle to Register and enter your Full Name, Email, Password, and your Company/Organization Name (this creates your isolated tenant space).
  4. Click Create Account. You will receive an email verification link.
  5. Workspace Provisioning: Upon successful registration, the backend automatically initializes a default tenant database schema, provisions unique widget API keys, and sets up your default agent profiles (Monica, Marketer, Business Finder).

2. Main Navigation & Modules

The platform is organized into distinct panels on the left-hand navigation bar:

[Overview]        - High-level health, telemetry, and pending task summary
[Inbox]           - The Human-in-the-Loop approval queue for agent actions
[Agent Fleet]     - Configuration panels for Monica, Marketer, Finder, etc.
[Memory Bank]     - Document upload center for RAG semantic knowledge ingestion
[Command Center]  - Real-time telemetry, intent stream monitor, and overrides
[Integrations]    - Connect API keys and accounts (Stripe, Apollo, Coinbase)
[Governance]      - Procurement audit logs, vendor verification, and invoice audits
[Finance]         - Financial runways, budget forecasts, and cashflow charts
[Settings]        - Workspace settings, domain locking, and profile details

3. Core Workflows for End-Users

Workflow A: Uploading Knowledge to the Memory Bank

Agents cannot perform context-aware tasks without data. The Memory Bank is your central RAG (Retrieval-Augmented Generation) ingestion engine. 1. Click on Memory Bank in the sidebar. 2. Drag and drop or upload reference files: .txt, .pdf, .docx, or .csv (e.g., your business FAQ, service terms, pricing sheets, or company templates). 3. The platform automatically: - Extracts and structures the text. - Chunks the text with overlapping context. - Generates vector embeddings using the Hugging Face connector. - Indexes chunks in the pgvector database and syncs them to the unified BrainBox Memory Substrate. 4. Once indexed, the status will show Vector Indexed (Green). Your agents will now automatically reference this content in their prompt context window.

Workflow B: Managing and Integrating Monica (CS Agent)

Monica is your customer service agent who runs via a chat widget on your public website. 1. Go to Agent Fleet > Monica. 2. Operational Mode Configuration: - Manual: Monica drafts answers, but every single response requires human approval. - Hybrid (Recommended): Monica answers common queries autonomously. If confidence is low or a query is high-risk (e.g., refunds), the action is sent to the Inbox for approval. - Autonomous: Monica runs fully on autopilot, resolving all queries and escalations without manual check-ins. 3. Behavior Customization: Edit the Personality Prompt text area (e.g., "You are a helpful customer support representative for Heyfae. Keep responses brief and polite."). 4. Website Widget Ingestion: - Copy the integration script tag from the panel: html <script src="https://api.agents.pluggedspace.org/static/accounts/js/monica-widget.js?api_key=YOUR_API_KEY" async></script> - Paste it before the closing </body> tag on your website. Monica is now active and domain-locked to your website.

Workflow C: Handling Action Approvals (The Inbox)

When an agent (like Monica or the Governance agent) triggers a critical action in Manual or Hybrid mode: 1. Navigate to Inbox. 2. Select any item from the left-hand feed. Each item details: - Agent: The agent that proposed the action (e.g., Monica). - Proposed Action: E.g., refund_invoice, send_email, or process_data. - Confidence Score: AI's self-assessed confidence (0.0 to 1.0). - Risk Level: Low, Medium, or High. - AI Reasoning: A concise explanation explaining why the agent chose this path. 3. Review the proposal, then click Approve (executes the payload and closes the task) or Reject (aborts the action).

Workflow D: Managing Integrations

Enable your agents to interact with the real world by connecting your software stack. 1. Click Integrations. 2. Locate the service provider you wish to connect (e.g., Stripe for billing, Apollo for lead scraping, or Coinbase for cryptocurrency checkout). 3. Enter your API credentials or click Connect Account to complete the standard OAuth handshake. 4. Once marked Connected, agents can safely query these integrations to check invoice statuses, fetch new sales leads, or process payments.

Workflow E: Monitoring Command Center & Logs

Track what your autonomous fleet is doing in real-time. 1. Command Center: View active tasks running in Celery worker pools, observe execution latency, and use the Emergency Stop override to temporarily pause all automated agent processes. 2. Execution Logs: Access the live stdout/stderr telemetry streams of background agent tasks (e.g., Marketer publishing posts or Business Finder scraping competitors) for debugging and audit logs.