Agents Overview
Understand AI agents in ThinkFleet — what they are, how they work, and when to use them.
Agents Overview
Agents are AI-powered assistants that can reason about tasks, use tools, and interact with users through natural conversation. While flows handle deterministic, event-driven automation, agents handle tasks that require judgment, planning, and dynamic decision-making.
What is an Agent?
An agent in ThinkFleet combines a large language model (LLM) with tools, memory, and a persona to create an assistant that can:
- Converse with users in natural language
- Reason about what actions to take
- Use tools to interact with external services
- Remember previous conversations and user preferences
- Search a knowledge base for relevant information
Agents vs Flows
| Feature | Flows | Agents |
|---|---|---|
| Trigger | Events (webhook, schedule, app) | User messages |
| Logic | Deterministic, predefined steps | LLM-driven, dynamic |
| Branching | Explicit conditions | AI decides based on context |
| Best for | Predictable, repeatable tasks | Open-ended, judgment-based tasks |
| Execution | Runs to completion | Iterative conversation |
In practice, agents and flows complement each other. Agents can invoke flows as tools, and flows can trigger agent conversations.
Agent Types
ThinkFleet supports two agent types:
Chatbot
A Chatbot is a conversational assistant focused on answering questions and providing information. Chatbots are ideal for:
- Customer support widgets
- Internal knowledge assistants
- FAQ bots
- Document Q&A
Chatbots typically rely on a knowledge base and have limited tool access.
Agent
An Agent is a more capable assistant that can plan, use multiple tools, and take multi-step actions. Agents are ideal for:
- Task automation (e.g., "Create a Jira ticket for this bug and notify the team")
- Data analysis (e.g., "Pull last month's sales data and summarize trends")
- Workflow orchestration (e.g., "Onboard this new customer across all our systems")
- Multi-step research (e.g., "Find all open invoices over 90 days and draft follow-up emails")
How Agent Execution Works
When a user sends a message to an agent:
1. User message arrives
2. System prompt + tools + memory are assembled
3. LLM processes the message
4. If the LLM calls a tool:
a. ThinkFleet executes the tool
b. The result is sent back to the LLM
c. The LLM may call more tools or generate a response
5. Final response is returned to the user
6. Conversation is stored in memory
This loop can repeat multiple times in a single turn. An agent might call 5-10 tools before formulating its final response, depending on the complexity of the request.
Tool Selection
The LLM automatically decides which tools to use based on:
- The user's request
- The tool descriptions and parameters
- The system prompt instructions
- Previous conversation context
You don't need to write branching logic or decision trees. The LLM handles routing.
Agent Components
System Prompt
The system prompt defines the agent's behavior, personality, and rules. It's the most important configuration for controlling agent quality.
You are a helpful customer support agent for Acme Corp.
Rules:
- Always greet the customer by name
- Check the knowledge base before answering product questions
- Escalate billing issues to the billing team
- Never share internal pricing formulas
- Be concise but thorough
Tools
Tools are the actions an agent can take. ThinkFleet supports multiple tool sources:
| Source | Description |
|---|---|
| Flow tools | Your ThinkFleet flows, exposed as callable tools |
| Piece tools | Individual piece actions (e.g., "Send Slack message") |
| Built-in MCP tools | Pre-built services like email, web scraping, database queries |
| External MCP servers | Third-party MCP-compatible tool servers |
Knowledge Base
Agents can search uploaded documents to ground their responses in your data. See Knowledge Base Overview for details.
Memory
Each agent maintains per-user conversation history and can use semantic search to recall relevant past interactions. See Agent Memory for details.
Persona
A persona is a preset configuration that bundles a system prompt, tool selection, and behavioral profile. ThinkFleet includes 20+ system personas for common roles (Developer, QA, Accountant, Research, etc.).
Deployment Options
Chat Widget
Embed an agent as a chat widget on your website:
<script src="https://your-instance.thinkfleet.com/chat/widget.js"
data-agent-id="agent_abc123"
data-theme="light">
</script>
API
Interact with agents programmatically via the REST API:
curl -X POST https://your-instance.thinkfleet.com/api/v1/agents/{agentId}/chat \
-H "Authorization: Bearer {apiKey}" \
-H "Content-Type: application/json" \
-d '{"message": "What are my open tickets?"}'
ThinkFleet Dashboard
Use agents directly from the ThinkFleet UI in the Tasks section.
Use Cases
| Use Case | Agent Configuration |
|---|---|
| Customer Support | Knowledge base + ticketing tools + escalation flows |
| Sales Assistant | CRM tools + email tools + knowledge base |
| DevOps Bot | GitHub + Jira + monitoring tools + deployment flows |
| Data Analyst | Database query tools + charting tools + export flows |
| HR Onboarding | HRIS tools + email templates + checklist flows |