Cost

How do I keep agent costs under control?

An operations manager looks at a usage chart while a small AI agent points out where to improve and other agents work in the background

Agents use credits every time they think and act. Most surprises come from three choices: how often an agent runs, how much it reads and which model it uses. All three are in your hands.

The challenge

An agent checks a mailbox every five minutes. A chat grows long. The most powerful model sorts simple forms. Each choice is small, but together they add up quietly. You need limits that actually hold and a clear view of where the credits go.

A short story

Lars runs operations and is an admin. In Agent usage, the list of all his agents, one agent stands out: it uses far more credits than the rest.

He asks it: "How can I reduce your credit usage?" The agent suggests three things: check the inbox every hour instead of every five minutes, move a long price list out of its instructions into a document it only reads when needed, and use a cheaper model for scheduled work while keeping the stronger model for chat.

Lars also sets a monthly limit for the agent. When it reaches 80% of the limit, a warning shows. If it reaches the limit, it pauses until next month or until an admin raises the limit. All its data and settings stay intact.

A month later, Lars checks Agent usage again to see what the changes did.

How does Abundly keep costs under control?

  • 1

    Set limits

    Limits stop the agent. They don’t just warn you.

    • Daily and monthly limits per agent, and monthly limits per team. Admins set them.
    • Personal agents share one monthly limit per person.
    • A monthly limit is a hard cap. A daily limit is a safety net: people with Edit access to the agent can reset it for the day.
    • A warning at 80%. At the limit the agent stops and all its data stays intact. If a chat was interrupted, a Continue button picks it up again once the limit is raised or reset.
  • 2

    Use the right model for the job

    Not every task needs the most powerful model.

    • Model selection with a clear cost indicator for each model.
    • Use one model for chat and a cheaper one for scheduled tasks or email.
    • Choose an effort level: Faster, Balanced or Smarter.
  • 3

    Design lean agents

    Small habits, big savings.

    • Run tasks only as often as needed. See agent optimization.
    • Keep reference material in documents the agent reads only when relevant, and keep instructions short.
    • Start a new chat when the context indicator in the chat turns orange or red.
    • Ask the agent itself how to lower its credit usage.
  • 4

    See where credits go

    Credits used next to value, per agent.

    • Agent usage shows each agent’s credits used next to its key metrics.
    • Admins see the credit balance, how fast it is used and its history over time.
    • Team admins get a Usage tab for each team they manage: credits used per day, week or month, agent by agent, ready to export.

Can admins control cost for the whole workspace?

  • 5

    Set the defaults

    Decide once, for everyone.

    • Choose the default model for new agents, switch models on or off, and require approval for new models.
    • A default daily limit for the workspace covers every agent that doesn’t have its own limit.
  • 6

    Adjust many agents at once

    No need to open each agent.

    • In Agent management, change model and limits for many agents in one go.
    • Or let a Workspace Manager agent set agent and team limits for you.
    • Keep credits in reserve for agents in production with lifecycle phases (Enterprise plan). Test agents pause first when the balance gets low.

Where do I start?

  1. 1

    Look at the top three

    Open Agent usage and find the agents that use the most credits.

  2. 2

    Set a monthly limit

    Give each important agent a cap that matches its value, or ask your admin to.

  3. 3

    Ask the agent

    "How can I reduce your credit usage?" Then try a cheaper model where quality allows.

See it in your own workspace

We will walk you through the platform with your own agents and rules in mind.

Request a demoMore questions answered