Creating Agents

Creating an agent gives you a dedicated AI persona for a specific area of your work.

Create a new agent

1. From your workspace home, click Create Agent.
2. Provide a name that reflects the agent’s role (e.g., “Investor Relations”, “Engineering Updates”, “Deal Monitor”).
3. Optionally add a description explaining what the agent is for.
4. Configure the agent’s system prompt — this defines the agent’s persona and how it approaches tasks.
5. Optionally configure the agent’s model to balance speed and capability.

Configuring an agent

After creating an agent, you can adjust its settings at any time:

• System prompt — Update the agent’s persona and instructions. This affects how the agent approaches all its responsibilities.
• Model — Optionally override which AI model the agent uses (Sonnet, Opus, or Haiku) to balance speed and capability.
• Members — Add team members to the agent. Members can see all of the agent’s execution sessions, including past ones.
• Security — A read-only audit of all tool permissions across the agent’s responsibilities. Tools are organized into two groups — Alma (built-in tools) and Integrations (connected third-party tools) — matching the layout used when adding tools. Within each toolkit card, tools are further separated into read and write sub-groups, each showing its configured approval level. A warning banner highlights any write tools that are set to auto-execute at medium or high risk, so you can quickly identify where human oversight may be warranted. Disabled tools are collapsed into a compact summary rather than listed in full.

Agent analytics

Each agent has an analytics page showing how active it is and what it costs. Navigate to an agent and click the Analytics tab to see:

Tasks view

The Tasks view shows how often the agent’s responsibilities run over a selected time period:

• A stacked bar chart broken down by responsibility.
• Total task count, average tasks per day, and a trend comparison to the previous period.
• A per-responsibility breakdown table with task counts and averages.

Costs view

The Costs view shows AI model usage and spend:

• A stacked cost bar chart by responsibility.
• Total cost, average cost per day, and a trend comparison.
• A per-responsibility cost breakdown table.

Switch between the two views using the Tasks / Costs toggle at the top of the analytics page.

Training

The Training section for each agent lets you review and refine how the agent learns from interactions. It includes a Learned Patterns tab that displays patterns Alma has inferred during training — things like extraction hints and topic associations. You can view, edit, or delete individual patterns inline.

Tips

• Keep agents focused — Each agent should represent a single role or domain. Create separate agents for deals, engineering, hiring, etc. rather than one agent that does everything.
• Start simple — Begin with one responsibility per agent and add more as you learn what works.
• Use descriptive names — Clear names help your team understand what each agent does at a glance on the home page.