Agents

Agents are AI-powered teammates that work on your behalf. They understand context, communicate across channels, and handle routine work automatically — while always keeping you in control.

What is an Agent?

An agent is:

  • An AI-powered role: Like having a digital staff member
  • Context-aware: Understands emails, tasks, and conversations
  • Action-oriented: Can send emails, create tasks, schedule meetings
  • Collaborative: Works with other agents and your team
  • Controlled: Always defers to you when needed (Human-in-the-Loop)

Agents can perform real work: schedule meetings, send replies, build reports, coordinate projects, and more.

How Agents Work

1. They Understand Context

Agents read and understand:

  • Emails: Content, sender, urgency, intent
  • Tasks: What needs to be done, deadlines, priorities
  • Conversations: Thread history, relationships, context
  • Metadata: Timestamps, attachments, routing information
  • Verified Senders: Using DKIM and other verification to know when you're interacting with them

2. They Make Decisions

Based on context, agents decide:

  • What to do: Reply, create task, escalate, delegate, or ignore
  • How to respond: Tone, content, urgency
  • When to ask: Escalate to you for approval (HITL)
  • Who to involve: Route to other agents or team members
  • Action alignment: Whether an action aligns with your identity or requires escalation

Agents are smart enough to understand when an action is aligned with your identity and when it's not, and will alert the HITL process when appropriate.

3. They Take Action (Safely)

Agents can:

  • Send emails on your behalf (with HITL approval for risky actions)
  • Create tasks from actionable items
  • Schedule meetings and manage calendar
  • Route messages to appropriate recipients
  • Coordinate with other agents
  • Generate content like reports or summaries
  • Ignore inbound communications when appropriate

Important: Agents never take risky actions without explicit approval via the HITL process. They are fault tolerant and rate limit constrained, so inbound emails or communications do not adversely impact token usage or execution limits.

4. They Learn

Agents improve over time:

  • From your feedback: Corrections and approvals
  • From patterns: What works, what doesn't
  • From context: Better understanding of your preferences
  • From policies: Following your rules and guardrails

Agent Types

Inbox Manager

Purpose: Manages your email inbox

Default Behavior:

  • Automatically created when you claim your Solo Identity
  • Active by default — handles all inbound email communications automatically
  • Can be disabled or deleted in the Team area if you don't want automatic inbox management

Capabilities:

  • Reads and categorizes emails
  • Drafts replies for your review
  • Creates tasks from actionable emails
  • Routes messages appropriately
  • Handles all inbound email communications actively (when enabled)

Assistant

Purpose: General help and task management

  • Handles routine questions
  • Manages your task list
  • Provides information and answers
  • Coordinates with other agents

Chief of Staff

Purpose: High-level coordination

  • Manages priorities and deadlines
  • Coordinates between agents
  • Handles strategic decisions
  • Provides executive support

Custom Agents

Create agents for specific purposes:

  • Billing Coordinator: Handle billing inquiries
  • Customer Success: Manage customer relationships
  • Marketing Coordinator: Handle marketing tasks
  • Project Manager: Coordinate projects

See Agent Types for complete list.

Agent Configuration

Each agent can be configured in multiple areas:

Basics

  • Name: Display name (e.g., "Jordan")
  • Kind: Agent type (e.g., "inbox-manager")
  • Status: Active or paused
  • Avatar: Emoji or image

Persona & Tone

  • Tone: Professional, friendly, formal, casual
  • Traits: Concise, empathetic, decisive, etc.
  • Signature: Email signature template
  • Style: Writing preferences

Working Patterns

  • Timezone: Agent's timezone
  • Working Hours: When agent is active
  • SLAs: Response time commitments
  • Off-hours: Behavior when inactive

Human-in-the-Loop (HITL)

  • Confidence Threshold: When to escalate
  • Intent Triggers: Always escalate for specific topics
  • Approval Mode: Hold for approval vs. send unless declined
  • Escalation Routes: Who to notify and when

Context & Knowledge

  • Datasets: Knowledge bases to reference
  • Retrieval Settings: How to find information
  • Refresh Schedule: When to update knowledge

Specialists

  • Allowed Specialists: Which specialists this agent can invoke
  • Priority Order: Which specialist to try first
  • Intent Mapping: Map intents to specialists
  • Auto-selection: Let system choose specialist

Policies & Guardrails

  • Policy Packs: Rules the agent must follow
  • Enforcement Level: Block, require approval, or warn
  • Overrides: Agent-specific policy adjustments

Tools & Integrations

  • Email: Provider and folder settings
  • Calendar: Provider and buffer settings
  • Storage: File storage providers
  • HTTP: Allowed hosts and methods

Model & Execution

  • Provider: Solo, OpenAI, Anthropic, Custom
  • Model: Specific model to use
  • Parameters: Temperature, max tokens, etc.
  • Budget: Daily/monthly spending limits

See Configuring Agents for details.

Agent Lifecycle

Creation

  1. Choose agent type
  2. Set basic information (name, kind)
  3. Configure settings
  4. Activate agent

Operation

  1. Agent receives context (email, task, etc.)
  2. Agent analyzes and decides action
  3. Agent takes action or escalates
  4. You review and provide feedback

Improvement

  1. Review agent actions
  2. Provide feedback and corrections
  3. Adjust configuration
  4. Agent learns and improves

Management

  • Pause: Temporarily disable agent
  • Update: Modify configuration
  • Delete: Remove agent (soft delete)
  • Clone: Duplicate agent with same settings

Agent Collaboration

Agents can work together:

  • Delegation: One agent delegates to another
  • Coordination: Agents coordinate on complex tasks
  • Specialists: Agents invoke specialists for specific tasks
  • Teamwork: Multiple agents handle different aspects

Human-in-the-Loop

Agents always keep you in control:

  • Never take risky actions without explicit approval
  • Escalate when uncertain, when confidence is low, or when actions don't align with your identity
  • Request approval for important actions
  • Hold drafts for your review before sending
  • Notify you of significant events or decisions
  • Alert you when actions may not be appropriate

Agents use verified sender technology (DKIM, etc.) to know when you're interacting with them, but they always require explicit approval for risky actions. You can configure when and how agents involve you. See Human-in-the-Loop for details.

Visibility & Control

You have complete visibility into all agent activity:

  • All inbound and outbound exchanges are visible via your SoloMail address
  • Configure agents to act differently based on your preferences
  • Set rules for how agents handle specific inbound emails or requests
  • Monitor activity to ensure agents are working as expected

If you don't want something handled by agents, you can:

  • Use a different email address for that communication
  • Configure agents to ignore or handle specific inbound emails differently
  • Set up rules in agent configuration to control behavior

Agent Limits

Agents are subject to plan limits:

  • Token Usage: AI tokens consumed per month
  • Executions: Number of agent runs per month
  • Storage: File attachments and data storage
  • Agents: Number of active agents (plan-dependent)

View your usage in Settings → Workflow.

Best Practices

Start Simple

  • Begin with default settings
  • Let agents learn from your interactions
  • Adjust configuration gradually

Set Clear Boundaries

  • Use HITL settings to control autonomy
  • Define escalation rules for sensitive topics
  • Set approval requirements for important actions

Monitor and Refine

  • Review agent actions regularly
  • Provide feedback when agents make mistakes
  • Refine configuration based on results

Use Specialists

  • Create specialists for specific tasks
  • Let agents invoke specialists when needed
  • Keep specialists focused and specialized

Apply Policies

  • Use policy packs for consistent rules
  • Set enforcement levels appropriately
  • Review policy violations regularly

Next Steps

Version: 1.0
Last Updated: November 2025