Plans + Pricing · Agents
Agents use AI credits, a monthly pool that is separate from your workflow runs. This article covers what credits are, how many your plan includes, how credits are calculated, and what happens when they run out.
What AI credits are
Every agent runs on a built-in AI model. You don't bring your own API key or need a separate AI subscription. Credits measure the work that model does for you.
Credits are a separate pool from workflow runs. Your workflow runs, limits, and rates don't change. Agents draw only from credits. Workflows draw only from runs.
What your plan includes
| Plan | AI credits per month | Published agents and workflows (combined) |
|---|---|---|
| Starter | 5,000 | 3 |
| Professional | 15,000 | 15 |
| Scale Up | 60,000 | 100 |
| Enterprise | Custom | Custom |
The published limit counts agents and workflows together. For example, if you're on Starter with 3 published workflows, you'll need to unpublish one or upgrade before you can publish an agent.
For plan prices and everything else included at each tier, see keragon.com/pricing.
How credits are calculated
In short: an agent thinks its way through each run, based on your instructions and the tools you've given it, instead of following a fixed script. The more it has to read, decide, and do, the more credits the run uses.
A workflow runs the same steps every time. An agent runs on a large language model (LLM), the kind of AI behind ChatGPT and Claude. It reads, decides the next step, calls a tool, and repeats until the job is done. Credits follow that work, so two runs of the same agent rarely cost the same.
What makes a run use more credits
- More to read. Long instructions, a 6-page referral PDF, or a long Spruce conversation thread all add to what the model processes.
- More steps. Every tool call means more to read and decide. An agent that checks eligibility, creates the patient in Athena Health, and posts to Slack #intake does more work per run than one that only sends a Slack message.
- More records. One agent that handles one new Healthie patient on Monday and a batch of 20 on Tuesday will use far more credits on Tuesday.
- Retries and loops. A tool that keeps failing, or an agent that can't tell when it's done, keeps reasoning and keeps using credits. See Troubleshooting agents.
Test runs use credits too, the same way live runs do.
What happens if you use all your credits
You'll get an email at 80% of your monthly credits. At 0, agents stop until your next billing cycle, and any run in progress shows "out of credits".
- Professional and Scale Up: turn on auto-recharge at the bottom of the Billing page to top up automatically. Top-ups are billed separately from your plan.
- Starter: capped at 5,000 credits, no auto-recharge. Upgrade to Professional to keep agents running.
- Enterprise: extra credits are billed at your custom rate.
Do credits roll over?
No. Your credit pool resets at the start of each billing cycle, and unused credits don't carry over to the next month.
Where to check your credit usage
Per run: the Runs tab. Open Runs and click the Agents tab. Each run shows the credits it used, next to its status and start time.
Your monthly total: Billing > Usage. The Agent credits tile shows how many credits you've used out of your monthly pool, the percentage used, and how many days until the pool refreshes.
By agent: Billing > Credits usage breakdown. See which agents use the most credits, split into live and test runs. Pick any date range.
Note: usage from before 14 October 2026 appears as Unattributed.
Keep your credit use in check
- Give the agent a clear finish line. Say what done means ("stop after the Slack message is posted") and how many times to retry before escalating. See Write your agent's instructions.
- Let workflows do the fixed steps. Use the agent for judgment, like deciding if a Spruce message is urgent, and a workflow for tagging, logging, and notifying. Workflows use runs, not credits. See Workflows, Agents, and MCP — what's the difference?
- Check usage weekly. In Billing > Credits usage breakdown, look for the agent using the most. Open its costliest runs: the same tool called again and again means the run looped. See Monitor your agent.
Questions
Not sure which plan fits your expected agent volume? Submit a request and our team will walk through it with you.