AI Agents

Build conversational AI agents powered by Claude, connected to your tools and integrations.

What Are AI Agents?

AI Agents in A91I are interactive assistants you can chat with in real time. Each agent is powered by Anthropic Claude and can call any of your connected integrations as tools — browsing Slack channels, querying a database, filing a Jira ticket, or sending an email, all within a single conversation.

Unlike workflows, which run in a fixed sequence from a trigger, agents reason dynamically. They decide which tools to call, in what order, and how to interpret results — all based on the conversation context and the system prompt you define.

Agents vs. Workflows

Use workflows for predictable, event-driven automation (e.g., "When a form is submitted, send a Slack message and create a Jira ticket"). Use agents for open-ended tasks that require reasoning, multi-step tool use, or natural language input from a user.

Key Concepts

System Prompt

The system prompt defines the agent's personality, role, and behavior. It tells the agent what it is, what it can do, and how it should respond. A well-written system prompt is the most important part of a useful agent.

  • Be specific about the agent's role (e.g., 'You are a customer support agent for Acme Corp').
  • List the tools available and when to use them.
  • Set the tone and response style.
  • Include any rules or constraints (e.g., 'Never share customer PII').

Conversation Threads

Each conversation with an agent is a thread. Threads maintain the full message history, so the agent can refer back to earlier context within the same conversation. You can start multiple threads with the same agent — each is isolated.

Agent Tools

Tools give agents the ability to take actions beyond just generating text. A91I supports two types of agent tools:

  • MCP Integration Tools — automatically available from your connected integrations (Slack, Gmail, Jira, GitHub, etc.).
  • Custom HTTP Tools — HTTP endpoints you define that the agent can call with structured parameters.

AI Model

Agents use Anthropic Claude. The model is configurable per agent (e.g., claude-sonnet-4-6 for high quality, claude-haiku-4-5 for speed). Claude 4.6 is the default and recommended choice for most use cases.

Agent Limits by Plan

The number of agents you can create depends on your subscription plan.

PlanMax AgentsAI Tokens
Builder (Free)31M tokens/month (shared)
Team10Unlimited (pay-as-you-go)
BusinessUnlimitedUnlimited (pay-as-you-go)
EnterpriseUnlimitedCustom dedicated pool

Agent AI usage (tokens consumed during conversations) draws from the same token budget as your workflow AI nodes. On the Team plan and above, AI is billed at cost with a 5% service fee. On the Builder plan, usage is capped at 1M tokens per month across both agents and workflows.

Next Steps

  • Building Your First Agent — create an agent with a system prompt and connected tools.
  • Custom HTTP Tools — extend agents with your own API endpoints.
  • Agents in Workflows — use an agent as a step inside a visual workflow.