Building Your First Agent
Create an AI agent with a system prompt, select a model, connect tools, and start chatting.
Creating an Agent
Navigate to Agents
Set a name and description
Write a system prompt
Select a model
Enable tools (optional)
Save and start chatting
System Prompt Tips
A good system prompt is specific, structured, and honest about what the agent can and cannot do.
Recommended Structure
- Role — Who is the agent? (e.g., 'You are a customer support agent for Acme Corp')
- Capabilities — What tools does it have and when should it use them?
- Constraints — What should it never do? (e.g., 'Never share internal pricing')
- Tone — How should it communicate? (e.g., 'Be concise and professional')
- Fallback — What to do when uncertain (e.g., 'If unsure, say so and ask for clarification')
Example System Prompt
Here is a starter prompt for a support triage agent:
You are a customer support triage agent for Acme Corp. Your job is to: 1. Understand the customer's issue from the conversation. 2. Search Slack for recent discussions about related topics. 3. Check Jira for any known bugs or open tickets. 4. Draft a helpful response for the support rep to send. Tools available: - Slack: search channels and read messages. - Jira: search issues, read details, add comments. Rules: - Never make promises about resolution timelines. - If you cannot find relevant info, say so honestly. - Keep responses under 200 words unless the issue requires detail.
Iterate your prompt
Model Selection
A91I agents run on Anthropic Claude. Choose the model based on the complexity of your use case and cost sensitivity.
| Model | Best For | Speed |
|---|---|---|
| claude-sonnet-4-6 (default) | Most tasks — balanced quality and speed | Fast |
| claude-opus-4-6 | Complex reasoning, long documents, nuanced tasks | Slower |
| claude-haiku-4-5 | Simple Q&A, high-volume, latency-sensitive | Fastest |
You can change the model in agent settings at any time. The model used is shown on each agent card.
Managing Conversations
Starting & Switching Threads
Each time you click New Thread inside the agent chat, you start a fresh conversation. The agent does not carry memory between threads — each thread is independent. Use threads to separate distinct tasks or client interactions.
Conversation History
All threads are saved and accessible from the thread list on the left side of the agent chat view. You can revisit any previous conversation to review the agent's reasoning, tool calls, and results.
Seeing Tool Calls
When the agent uses a tool, the chat shows a collapsible tool call block with:
- The tool name and which integration it belongs to.
- The input parameters the agent sent.
- The raw result returned by the tool.
This makes it easy to audit what the agent did and debug unexpected behavior.
Agent Settings
Click the settings icon on any agent card to open the settings page. From there you can:
- Rename the agent or update its description.
- Edit the system prompt.
- Switch the AI model.
- Enable or disable individual integration tools.
- Add or remove custom HTTP tools.
- Delete the agent (this also deletes all conversation history).