Agents ask more questions than people do
For most of the history of observability, the number of questions a team could ask of its data was limited by how many people had time to ask them. Agents remove that limit. They can query all day and all night, many at once, so how much a company investigates no longer depends on the size of its on-call rotation.
That raises a new question. How much should each person and each agent be able to query, and who gets to decide?
Query budgets are our answer. You can now give any API token or role an hourly budget, a daily budget, or both, in dollars. They're available on Axiom Cloud at no extra cost. In Console, you'll find them under query cost limits.
Agents are doing more of the looking
Agents already work in a lot of places. An engineer asks Claude, Claude Code, or Cursor a question, and the agent queries Axiom through our MCP Server with that engineer's access. A scheduled job compares this morning's error rates with last week's and reports only what changed. An AI teammate sits in a team's channel and answers whoever asks.
The more of this a company does, the earlier it finds problems, because agents keep looking when nobody else is. The companies that get the most from them let agents check every deploy, alert, and customer report as it happens, and spend their people's time deciding what to do about what the agents find.
That only works if agents see everything a person would. Budgets limit querying, never ingest, so the data an agent reads at 3am is just as complete as what an engineer reads in the morning.
Every query runs as a person or a token
A budget belongs to whoever is running the query, and in Axiom that's either a person or a token.
People. When you connect an agent through the Axiom MCP Server, it acts as you. Its queries come out of the same budget as your queries in Console and your personal access tokens. That budget comes from your role, and each person on the role gets their own. A $20 daily budget on a role gives every member $20 a day.
Tokens. An agent that runs on its own, like a scheduled job, a CI script, or a service, gets its own API token with only the permissions it needs and its own budget. You can add a budget to a token at any time without regenerating it, so agents already in production can have one today.
A token's budget is also its worst case. A token with a $10 daily budget can't spend more than $10 a day on queries, whatever it decides to run.
A thousand people, organized by team
Picture a company of a thousand people. One way to set it up is to give each team a group: Security, Infrastructure, Support, Product. Each group's role sets what the team can do and how much it can query each day. When someone changes team, their budget changes with them, because it was never set on them personally.
If someone has more than one role, the smallest budget wins, so joining a new group can never raise what a person can spend by accident. Roles you don't want to limit can simply have no budget.
On Axiom Cloud, you can set budgets on API tokens and on the five built-in roles. With the RBAC add-on, you can set them on custom roles and groups too.
Agents can see their budget
A budget works best when the agent knows about it. Every query an agent runs through MCP comes back with what it cost and how much budget is left. An agent that has spent most of its hour on one broad search can make the next one narrower.
When a budget runs out, only that token or person is stopped, and they're told which budget ran out and when it resets. Everyone else keeps working, and so does every dashboard and monitor. Hourly budgets reset at the top of each hour and daily budgets at midnight, both in UTC.
In Console, you can watch usage against each budget on the token page, in the Users list, and on each person's profile.
Budgets cover queries, the spending limit covers the bill
Budgets only ever limit querying. A token or person keeps sending data no matter how much of its query budget it has used.
Above every budget sits your organization's monthly spending limit, which covers the whole bill, ingest included. If you want one firm ceiling for the month, that's the control. Budgets decide how the querying underneath it is shared out.
Let the agents in
Open Settings > API tokens and give your next agent a budget of its own. Then set budgets on your roles, so the next team that wants to put agents on its data can get a yes the same day.
We built Axiom so you could keep all your machine data and ask it anything. Now agents are asking too, and they'll ask far more questions than people ever had time to. That's good news for anyone who runs software. Budgets make it easy to say yes, and we can't wait to see what your agents find.