AI Agents are autonomous workers that take real actions in your workspace on a schedule or in response to events — without you having to be in the dashboard. They are distinct from Artemis (which responds to your chat) and from Workflows (which execute fixed steps): agents have their own decision logic, accounts, and approval queue.
The three agent types
1. Mention Reply Agent
Watches your Mentions feed and drafts replies to relevant mentions across X, Reddit, LinkedIn, and other sources. You configure the topics it should respond to, the tone, and which accounts it can post from. Replies enter the Approval queue by default so you can review before they go public; you can switch to auto-post once you trust it.
2. Lead Collector Agent
Automates ongoing lead acquisition. Tell it your ideal customer profile (industry, role, region, signals) and it runs LinkedIn searches, Sales Navigator queries, and Instagram lookups on a recurring schedule. Leads land in a target list, deduplicated and optionally enriched. Replaces the manual run Lead Finder every Monday loop.
3. Conversation Reply Agent
Drafts replies to incoming conversations in Unibox — campaign replies, social DMs, ad-hoc messages. Uses your workspace context, brand voice, and historical replies as training. Approvals queue lets you review before sending; auto-send is available for trusted scenarios.
Creating an agent
Open AI Agents → + New agent.
Pick the agent type.
Configure its instructions: topics, audience, tone, constraints. Artemis can draft these for you.
Assign sending accounts — which inboxes or social profiles the agent is allowed to use.
Pick the approval mode: Review every action, Auto-approve trusted patterns, or Fully autonomous.
Activate. The agent runs immediately and on its schedule.
The approval queue
Every agent action — a drafted reply, a new lead added to your CRM, a scheduled post — enters the approval queue first by default. You can:
Approve and send.
Edit and send.
Reject (the agent learns from the rejection).
Approve and mark as a pattern (so similar actions auto-approve in future).
Pending approvals show a badge in the sidebar so you do not miss them. Email and Slack notifications can mirror that badge if you do not check the dashboard daily.
Activity log
Every agent has a per-agent activity log showing what it has done, what is pending, and what was rejected. Click into any row to replay the agent reasoning for that action.
Agent settings
Schedule: how often the agent wakes up and looks for work.
Account limits: max actions per account per day so you do not blow through provider rate limits.
Brand voice: link to your workspace context so the agent matches tone.
Stop conditions: pause the agent if it accumulates a certain number of rejections in a window.
How agents differ from workflows
Workflows execute a deterministic sequence of steps based on events. They do exactly what you build.
Agents use AI to decide what to do within a goal you set. They handle ambiguity, draft content, and adapt over time.
You can also wire an agent into a workflow as a node — for example, a workflow that triggers on a new mention can route it to the Mention Reply Agent for a draft, then send the draft into the approval queue.
Costs
Agents consume AI credits per action. The credit cost is shown on the agent detail page and aggregated in Settings → Billing → Usage. BYOK users pay their provider directly and avoid plan credits — see the BYOK article.
Plan availability
Mention Reply Agent: Pro and above (SaaS); AppSumo Tier 2+.
Lead Collector Agent: Pro and above; AppSumo Tier 2+.
Conversation Reply Agent: Growth and Agency; AppSumo Tier 3+.