AI Agent Marketplace

Overview

The AI Agent Marketplace in SalesOS is a curated catalog of pre-built intelligent agents that automate specific sales workflows, tasks, and processes. Each agent operates autonomously within defined boundaries, handling repetitive work so your team can focus on high-value selling activities.

Agents range from simple task automators (scheduling follow-ups) to sophisticated AI-powered systems (qualifying inbound leads through multi-turn conversations). The marketplace model allows your organization to adopt automation incrementally, installing only the agents that match your current needs and expanding over time as comfort and trust grow.

What is an AI Agent?

An AI agent in SalesOS is an autonomous software component that:

• Observes specific triggers or conditions in your CRM data.
• Decides what action to take based on configured rules and AI reasoning.
• Acts by creating, updating, or communicating within defined boundaries.
• Learns from outcomes to improve performance over time.

Unlike simple automation rules (if X then Y), agents can handle ambiguity, make judgment calls within parameters, and adapt their behavior based on context and historical patterns.

Agent vs. Automation vs. Workflow

Browsing the Marketplace

Access the marketplace from Settings > AI Agents > Marketplace or the dedicated Agents tab in the main navigation.

Marketplace Layout

The marketplace presents agents in a browsable catalog with:

• Featured agents: Curated top picks displayed prominently.
• Category navigation: Browse by functional area.
• Search: Find agents by name, capability, or use case keyword.
• Filters: Narrow by category, rating, install count, or publisher.
• Sort options: Popular, highest rated, newest, or alphabetical.

Agent Categories

Agent Cards

Each agent in the marketplace is presented as a card containing key information for evaluation.

Card Information

Agent Detail Page

Clicking an agent card opens its full detail page with:

• Extended description: Detailed explanation of capabilities and use cases.
• Screenshots/demos: Visual examples of the agent in action.
• Capabilities list: Specific actions the agent can perform.
• Data access requirements: What CRM data the agent needs to read or write.
• Integration requirements: External services or APIs the agent connects to.
• Configuration options: Preview of available settings.
• Reviews and ratings: User feedback from other organizations.
• Changelog: Version history with feature additions and bug fixes.
• Publisher information: Company details, support contact, documentation links.

Installing Agents

Installation Process

1. Select Agent: Click Install on the agent card or detail page.
2. Review Permissions: Examine what data access and actions the agent requires.
3. Accept Terms: Acknowledge the agent's terms of service and data handling policy.
4. Configure Basics: Set initial configuration (can be refined later).
5. Activate: Choose to activate immediately or install in inactive state for later configuration.

Installation Requirements

Before installation, SalesOS verifies:

• Your plan includes access to the agent (or premium add-on is purchased).
• Required integrations are connected (e.g., email integration for email-based agents).
• Required data exists (e.g., opportunity data for deal coaching agents).
• User has admin or agent-manager permissions.

Permissions and Data Access

Permission Model

Each agent declares its required permissions using a granular access model:

Data Access Scopes

Agents operate within defined data boundaries:

• Object scope: Which CRM objects the agent can access (e.g., Leads and Contacts only).
• Field scope: Which specific fields within those objects (e.g., email, phone, company, but not internal notes).
• Record scope: Which records (e.g., only records owned by users who opted in, or only records in specific pipeline stages).
• Temporal scope: How far back in history the agent can look (e.g., last 90 days of activities).

Security Controls

• Principle of least privilege: Agents only receive the minimum permissions necessary.
• Audit logging: Every action taken by an agent is logged in the audit trail.
• Data isolation: Agents cannot access data outside their declared scope.
• Encryption: Agent-stored data (caches, learned patterns) is encrypted at rest.
• Revocation: Permissions can be revoked instantly, immediately halting agent activity.

Configuring Agent Behavior

After installation, agents are configured through a structured settings interface.

Trigger Configuration

Define when the agent activates:

Schedule Configuration

For schedule-based agents:

• Frequency: Hourly, daily, weekly, or custom cron expression.
• Time window: Restrict operation to business hours or specific time ranges.
• Timezone: Operate in the organization's timezone or per-rep timezone.
• Blackout periods: Suspend during holidays, freezes, or events.

Boundary Configuration

Define the limits of agent autonomy:

Behavior Tuning

Many agents support behavioral parameters:

• Aggressiveness: How proactively the agent takes action (conservative vs. assertive).
• Communication tone: Formal, conversational, or custom brand voice.
• Decision criteria: Weighted factors for scoring and prioritization.
• Learning rate: How quickly the agent adapts based on feedback.

Monitoring Agent Activity

Activity Dashboard

Each installed agent has a dedicated activity dashboard showing:

• Recent actions: Chronological log of what the agent did.
• Action breakdown: Pie chart of action types (created, updated, sent, skipped).
• Volume over time: Line chart showing daily/weekly action counts.
• Success rate: Percentage of actions that achieved desired outcomes.
• Errors and exceptions: Failed actions with error details.
• Escalations: Cases where the agent deferred to human judgment.

Activity Feed

The global agent activity feed aggregates actions across all installed agents:

[09:15] Lead Researcher: Enriched 23 new leads with company data
[09:30] Smart Follow-Up: Sent 8 follow-up emails for stale opportunities
[10:00] Deal Coach: Generated coaching suggestions for 5 at-risk deals
[10:15] Duplicate Finder: Identified 3 potential duplicate contacts for review
[11:00] Pipeline Narrator: Generated weekly pipeline summary report

Alerts and Notifications

Configure notifications for agent events:

• Error alerts: Immediate notification when an agent encounters repeated failures.
• Threshold alerts: Notification when agent activity exceeds normal patterns.
• Summary digests: Daily or weekly summaries of all agent activity.
• Escalation alerts: When agents need human input to proceed.

Performance Metrics

Measuring Agent Value

Each agent tracks performance metrics specific to its function:

ROI Calculation

SalesOS calculates estimated ROI for each agent:

• Time saved: (actions per period) x (average manual time per action) x (loaded rep cost per hour).
• Revenue impact: Deals influenced by agent actions x average deal value x win rate delta.
• Error reduction: Manual error rate vs. agent error rate x cost per error.

Benchmarking

Compare your agent performance against:

• Your own historical baseline: Before vs. after agent installation.
• Organization average: How your instance compares to similar-sized deployments.
• Category benchmark: Agent performance vs. others in the same category.

Custom Agent Creation

Organizations can build custom agents tailored to their unique workflows.

Agent Builder

The no-code Agent Builder provides a visual interface for creating custom agents:

1. Define purpose: Name, description, and category for your agent.
2. Set triggers: Configure when the agent should activate.
3. Define logic: Build decision trees, scoring models, or AI prompts.
4. Specify actions: What the agent does (create records, send messages, update fields).
5. Set boundaries: Limits, escalation rules, and safety constraints.
6. Test: Run the agent in simulation mode against historical data.
7. Deploy: Activate for production use.

AI Prompt Agents

For agents that leverage AI reasoning, you provide:

• System prompt: Define the agent's role, personality, and constraints.
• Context template: What data to provide the AI for each decision.
• Output schema: Expected response format and valid actions.
• Evaluation criteria: How to measure if the AI's decisions are correct.

Code-Based Agents

For advanced use cases, developers can build agents using the Agent SDK:

• TypeScript/JavaScript SDK with full type definitions.
• Event-driven architecture with typed triggers.
• Built-in state management and persistence.
• Testing framework for unit and integration tests.
• CI/CD pipeline integration for deployment.

Publishing to Org Marketplace

Custom agents can be published to your organization's private marketplace.

Publishing Process

1. Prepare: Ensure the agent is documented, tested, and handles edge cases.
2. Package: Bundle configuration, code, and metadata.
3. Review: Submit for internal review by an admin or designated reviewer.
4. Publish: Once approved, the agent appears in the org's private marketplace.
5. Maintain: Monitor adoption, collect feedback, and publish updates.

Publication Requirements

Managing Installed Agents

Agent Management Dashboard

The management dashboard (Settings > AI Agents > Installed) shows all active agents with:

• Status (active, paused, error, disabled).
• Last activity timestamp.
• Performance summary (actions this period, success rate).
• Quick actions (pause, configure, uninstall).

Lifecycle Management

Version Management

• Auto-update: Agent automatically updates to new versions (default for official agents).
• Manual update: Review changelog before accepting updates.
• Pin version: Lock to a specific version regardless of new releases.
• Rollback: Revert to a previous version if an update causes issues.

Multi-Agent Coordination

When multiple agents operate on the same data:

• Priority ordering: Define which agent takes precedence in conflicts.
• Mutual exclusion: Prevent multiple agents from modifying the same record simultaneously.
• Chaining: One agent's output can trigger another agent's input.
• Shared context: Agents can reference each other's recent actions for coordinated behavior.

Best Practices

1. Start with one agent and expand gradually. Install a single agent, observe its behavior for two weeks, and validate that it operates as expected before adding more. Rushing to install multiple agents creates complexity that is difficult to debug.

2. Set conservative boundaries initially. Begin with tight action limits and high confidence thresholds. It is easier to loosen boundaries once trust is established than to recover from an overly aggressive agent taking unwanted actions.

3. Monitor actively during the first week. Review the agent's activity feed daily during the initial period. Early observation catches misconfigurations before they compound into larger issues.

4. Use human-in-the-loop for high-stakes actions. For actions that affect customer relationships or involve significant financial decisions, configure agents to recommend rather than act autonomously. Let humans approve until the agent proves reliable.

5. Provide clear feedback on agent decisions. When reviewing escalations, mark agent suggestions as correct or incorrect. This feedback loop directly improves agent accuracy over time for agents with learning capabilities.

6. Review permissions regularly. As your usage evolves, agents may retain permissions they no longer need. Quarterly permission audits ensure agents maintain minimal required access.

7. Document custom agents thoroughly. Custom agents should have clear documentation explaining their purpose, logic, boundaries, and failure modes. When the original builder is unavailable, others need to understand and maintain the agent.

8. Coordinate agent interactions. Before installing a new agent, review whether it overlaps with existing agents. Define clear ownership boundaries to prevent duplicate actions or conflicting updates on the same records.

9. Measure ROI objectively. Track time saved and outcome improvements against a baseline period. Agents that do not demonstrably improve outcomes after a reasonable trial period should be reconfigured or retired.

10. Keep agents aligned with process changes. When your sales process, territories, or team structure changes, review all active agents for necessary updates. Agents operating on outdated assumptions produce incorrect actions.