Understanding AI Agents

AI agents are the autonomous workers of Solatis. They're intelligent systems that can perceive, decide, and act with minimal human intervention. This guide explains how they work and why they're transformative.

What is an AI Agent?

An AI agent is a software program that:

Perceives its environment (reads documents, analyzes data, monitors systems)
Reasons about what to do (uses AI to make decisions)
Acts to accomplish goals (creates tasks, sends messages, updates systems)
Learns from feedback (improves over time)
Operates autonomously (works 24/7 without human intervention)

Real-World Example

Without AI Agent:

1. Meeting ends at 3 PM
2. Someone transcribes (takes 30 min)
3. Someone reviews transcript (15 min)
4. Someone extracts action items (10 min)
5. Someone creates Jira tickets (15 min)
6. Someone sends summary email (5 min)
   Total: 1.5+ hours, manual work

With AI Agent:

1. Meeting ends at 3:05 PM
2. Agent automatically:
   - Transcribes meeting
   - Identifies action items
   - Creates Jira tickets
   - Assigns to people
   - Sends summary emails
   Total: 3 minutes, fully automated

How Agents Work: The Cycle

Phase 1: Perception 👁️

Agent observes triggers and events:

"A meeting just ended"
"New Slack messages in #product"
"Document uploaded to knowledge base"
"Scheduled time for daily briefing"

Phase 2: Planning 🧠

Agent reasons about the goal:

What needs to be done?
What's the best approach?
What tools do I need?
What's the sequence of steps?

Phase 3: Execution ⚙️

Agent takes actions:

Calls tools (transcribe, analyze, search)
Makes decisions
Creates outputs
Handles errors

Phase 4: Explanation 📋

Agent reports what happened:

What did I do?
Why did I make each decision?
What was the confidence level?
What sources did I use?

This cycle repeats continuously, improving with feedback.

Agent Types in Solatis

1. Task Agents (Bounded Execution)

Purpose: Execute specific, well-defined tasks

Characteristics:

10-step maximum
300-second timeout
50K token limit
Works on focused goals

Example: Contract Analysis Agent

Goal: Analyze contract and extract key terms

Step 1: Extract text from PDF
Step 2: Identify contract type (NDA, Service Agreement, etc.)
Step 3: Find key dates (signature, expiration, renewal)
Step 4: Extract financial terms (pricing, payment schedule)
Step 5: Identify termination clauses
Step 6: Flag unusual provisions
Step 7: Generate executive summary
Step 8: Recommend next actions

Output: Structured contract analysis

When to use:

One-time analysis tasks
Well-defined workflows
Document processing
Simple extractions

2. Conversational Agents (Interactive)

Purpose: Have natural conversations while maintaining context

Characteristics:

16K token context window
Long-term memory
Short-term memory
Conversation history
Multi-turn interactions

Example: Knowledge Assistant

User: "What's our pricing strategy?"
Agent: [Searches company docs, meetings, policies]
Output: "Based on our Q1 planning meeting..."

User: "Who owns that decision?"
Agent: [Uses previous context]
Output: "Sarah from Product leads pricing. You met about this on..."

User: "Send her my questions"
Agent: [Takes action]
Output: "Message sent to Sarah with your 3 questions"

When to use:

Answering questions
Exploring topics
Multi-step discussions
Knowledge discovery

3. Sentinel Agents (Always Watching)

Purpose: Monitor continuously and alert on important events

Characteristics:

50-step maximum
3600-second timeout (1 hour)
100K token limit
Runs on schedule
Proactive monitoring

Example: Team Health Sentinel

Runs every hour:
1. Check all open tasks and deadlines
2. Identify at-risk items (due in <24 hours)
3. Check team Slack for blockers
4. Analyze meeting trends
5. Identify skill gaps
6. Compile briefing
7. Alert manager if issues found

Outputs: Daily briefing, alerts

When to use:

Daily/weekly monitoring
Proactive issue detection
Regular status reports
Continuous oversight

4. Orchestrator Agents (Master Coordinator)

Purpose: Coordinate multiple agents working together

Characteristics:

Manages agent pools
Shared memory for all agents
Conflict resolution
Task distribution
Consensus building

Example: Company Strategy Agent

Coordinates:
- Market Analysis Agent (industry trends)
- Financial Agent (company health)
- Competitor Analysis Agent (competitor moves)
- Team Performance Agent (internal capacity)

Process:
1. All agents gather their data
2. Orchestrator synthesizes findings
3. Resolves any conflicts
4. Generates strategic briefing

Output: Executive strategy briefing

When to use:

Complex multi-step workflows
Need multiple perspectives
Coordinating teams
Strategic analysis

Agent Components

The Brain: Language Model

The agent's reasoning engine (powered by LLMs):

Model	Strengths	Best For
Claude 3.5 Sonnet (Default)	Reasoning, analysis, accuracy	Complex decisions, analysis
GPT-4	General intelligence, speed	Varied tasks, quick decisions
Claude 3 Opus	Deep reasoning, long context	Complex strategic analysis

The Tools: What Agents Can Do

Agents access tools to accomplish tasks:

Tool	What It Does
semantic_search	Find information across documents (by meaning)
document_chat	Ask questions about specific documents
transcribe	Convert audio/video to text
extract_entities	Find people, dates, topics in text
generate_summary	Create brief summaries
create_task	Make Jira/Asana tickets
send_slack_message	Communicate via Slack
check_calendar	Look up availability, schedule
analyze_sentiment	Understand tone and emotion
extract_numbers	Find metrics and data

The Memory: What Agents Remember

Agents maintain different types of memory:

Short-term Memory

Current conversation
Recent actions
Immediate context
~16K tokens

Long-term Memory

Learned patterns
User preferences
Historical decisions
Past interactions
Unlimited capacity

Shared Memory (for swarms)

Accessible by multiple agents
Coordination information
Conflict resolution data
Consensus tracking

Agent Configuration

Every agent in Solatis is configured with:

yaml

Name: Meeting Action Item Agent
Type: Task Agent
Description: Extract action items from meeting transcripts

Tools Available:
  - semantic_search
  - create_task
  - send_slack_message
  - extract_entities

Model: Claude 3.5 Sonnet
Max Steps: 10
Timeout: 300 seconds
Max Tokens: 50,000

Memory:
  Short-term: 16K tokens
  Long-term: Enabled
  Learning: Enabled

Security:
  Data Isolation: Workspace-level
  API Key Encryption: Yes
  Audit Logging: Yes
  Sandbox Mode: Yes

Built-in Agents

1. Meeting Analysis Agent

What it does: Automatically processes meeting recordings

Trigger: Meeting ends

Process:

Transcribe audio with 95%+ accuracy
Identify speakers
Extract action items with owners
Generate executive summary
Create Jira tickets for tasks
Send summary to participants
Store transcript in knowledge base

Output: Summary + Jira tickets + Email

2. Knowledge Retrieval Agent

What it does: Answer questions by finding and synthesizing information

Trigger: User asks a question

Process:

Understand what the user is asking
Search documents and meetings
Rank results by relevance
Synthesize information
Cite sources
Ask clarifying questions if needed

Output: Detailed answer with sources

3. Task Automation Agent

What it does: Create and manage tasks automatically

Trigger: Action items identified, deadlines approaching

Process:

Extract task details
Determine owner and priority
Calculate deadline
Create in Jira/Asana/Trello
Set reminders
Track completion

Output: Tasks in your tools

4. Content Analysis Agent

What it does: Analyze documents and extract insights

Trigger: Document uploaded

Process:

Extract text and structure
Identify document type
Extract named entities (people, organizations, dates)
Identify sentiment (positive/negative/neutral)
Generate summary
Create tags for organization
Link to related documents

Output: Analyzed document with metadata

5. Competitor Intelligence Agent

What it does: Monitor competitors and extract insights

Trigger: Scheduled (daily/weekly)

Process:

Search for competitor mentions
Analyze news and announcements
Extract key developments
Compare to our strategy
Identify threats and opportunities
Generate briefing

Output: Intelligence briefing

Agent Reliability & Safety

How Solatis Ensures Agents Work Correctly

1. Explainability

Every action is logged
Reasoning is documented
Decisions are traceable
Human-readable explanations

2. Approval Gates

High-risk actions require approval
Sensitive data access is logged
Cost-intensive operations have limits

3. Resource Limits

Maximum steps per agent
Timeout protection
Token spending limits
Rate limiting

4. Sandboxing

Agents run in isolated environments
Network access controlled
File access restricted
Data isolation enforced

5. Audit Trail

Everything logged for compliance
Full execution history
Decision tracking
Error documentation

Creating Custom Agents

You can create custom agents for your specific workflows.

Example: Lead Scoring Agent

yaml

Name: Lead Scoring Agent
Type: Task Agent

Trigger: New lead added to CRM

Process:
1. Extract lead information
2. Search for company info
3. Analyze email domain
4. Check website/LinkedIn
5. Score based on criteria
6. Add score to CRM
7. Notify sales team if high-score

Tools:
  - semantic_search
  - external_search
  - crm_update
  - slack_notify

See Building Custom Agents for detailed instructions.

Agent Performance

Typical Results

Task	Manual Time	Agent Time	Accuracy
Meeting summary	15 min	2 min	95%+
Jira ticket creation	10 min per ticket	<1 min per ticket	98%+
Document analysis	30 min	3 min	92%+
Action item extraction	20 min	1 min	96%+
Competitor research	1 hour	5 min	94%+

Cost Savings Example

Scenario: Engineering team of 10

Monthly Meetings: 100
Manual Processing:
- Transcription: 100 × 0.5 hours = 50 hours
- Summary: 100 × 0.25 hours = 25 hours
- Action items: 100 × 0.25 hours = 25 hours
Total: 100 hours/month = $5,000 (at $50/hr)

With Solatis Agents:
- Automation cost: ~$200/month
- Human time saved: 95+ hours/month
- Monthly savings: $4,800+

Best Practices

✅ Do's

✅ Start with simple agents
✅ Monitor first actions carefully
✅ Provide feedback to improve
✅ Use agents for repetitive tasks
✅ Set clear approval gates
✅ Review agent outputs regularly
✅ Document agent workflows

❌ Don'ts

❌ Give agents access to too many tools
❌ Skip monitoring initial deployments
❌ Ignore agent explanations
❌ Use agents for one-time tasks
❌ Assume perfection (always review)
❌ Remove human oversight completely
❌ Overload agents with complexity

Troubleshooting Agents

Agent Not Running

Check schedule/trigger is enabled
Verify agent has required permissions
Check tool access is configured
Review error logs

Wrong Output

Provide feedback to retrain
Clarify goal in agent config
Add more context/examples
Adjust prompts

Too Slow

Reduce steps/complexity
Add caching for searches
Parallelize agent swarms
Use faster model

High Cost

Reduce token usage
Use faster model
Limit frequency
Optimize queries

Next Steps

Building Custom Agents - Create your own workflows
Agent Security - How we keep agents safe
Troubleshooting Agents - Common issues and fixes
Contact Support - Get help from our team

Ready to automate? Start with pre-built agents, then customize for your needs! 🚀

Understanding AI Agents ​

What is an AI Agent? ​

Real-World Example ​

How Agents Work: The Cycle ​

Phase 1: Perception 👁️ ​

Phase 2: Planning 🧠 ​

Phase 3: Execution ⚙️ ​

Phase 4: Explanation 📋 ​

Agent Types in Solatis ​

1. Task Agents (Bounded Execution) ​

2. Conversational Agents (Interactive) ​

3. Sentinel Agents (Always Watching) ​

4. Orchestrator Agents (Master Coordinator) ​

Agent Components ​

The Brain: Language Model ​

The Tools: What Agents Can Do ​

The Memory: What Agents Remember ​

Agent Configuration ​

Built-in Agents ​

1. Meeting Analysis Agent ​

2. Knowledge Retrieval Agent ​

3. Task Automation Agent ​

4. Content Analysis Agent ​

5. Competitor Intelligence Agent ​

Agent Reliability & Safety ​

How Solatis Ensures Agents Work Correctly ​

Creating Custom Agents ​

Example: Lead Scoring Agent ​

Agent Performance ​

Typical Results ​

Cost Savings Example ​

Best Practices ​

✅ Do's ​

❌ Don'ts ​

Troubleshooting Agents ​

Agent Not Running ​

Wrong Output ​

Too Slow ​

High Cost ​

Next Steps ​

Understanding AI Agents

What is an AI Agent?

Real-World Example

How Agents Work: The Cycle

Phase 1: Perception 👁️

Phase 2: Planning 🧠

Phase 3: Execution ⚙️

Phase 4: Explanation 📋

Agent Types in Solatis

1. Task Agents (Bounded Execution)

2. Conversational Agents (Interactive)

3. Sentinel Agents (Always Watching)

4. Orchestrator Agents (Master Coordinator)

Agent Components

The Brain: Language Model

The Tools: What Agents Can Do

The Memory: What Agents Remember

Agent Configuration

Built-in Agents

1. Meeting Analysis Agent

2. Knowledge Retrieval Agent

3. Task Automation Agent

4. Content Analysis Agent

5. Competitor Intelligence Agent

Agent Reliability & Safety

How Solatis Ensures Agents Work Correctly

Creating Custom Agents

Example: Lead Scoring Agent

Agent Performance

Typical Results

Cost Savings Example

Best Practices

✅ Do's

❌ Don'ts

Troubleshooting Agents

Agent Not Running

Wrong Output

Too Slow

High Cost

Next Steps