Running agents & reading traces
A run is one execution of your project: an input goes in, each step happens in order, and every run is saved with its full trace.
Runs start three ways — you press Run and watch it live, a schedule fires with nobody watching, or the client's own system calls the project. All three take exactly the same path and leave the same kind of trace, which is deliberate: the ones nobody is watching are the ones you most need to be able to read afterwards.
Start a run
- Open a project and click Run in the toolbar.
- Type an input if the job needs one — a customer's message, a batch of notes, a question. (Some projects run fine with no input.)
- Click Run. Steps appear live as each agent finishes — you'll see the agent's role and its output, with a pulse while the next step is thinking.
- When the run completes, the full saved trace opens in the Execution tab.
Read a trace (Execution tab)
The Execution tab is the project's run history:
- Left: every past run — the input, when it ran, and whether it succeeded or failed. Scheduled runs and calls from a client's system appear here too, which is how you check that something you set up last month is still working.
- Right: the selected run in full — the input, each step (which agent, what it produced), the final output, and the error if something went wrong.
Traces are how you improve a project: find the step where things went sideways, fix that agent's goal or knowledge, and run again.
Run limits (quotas)
To keep the platform fast and costs sane, runs are limited per day:
| Account | Runs per day |
|---|---|
| Expert | 100 |
| Practice sandbox (Academy) | 10 |
Hit the limit and you'll see "Daily run limit reached" — it resets at midnight UTC. Model usage per run is also recorded; Claude runs are on RAiN, and runs on your own provider keys are billed by that provider.
Download your project
Download in the toolbar exports the whole solution as a JSON file — the agents, the steps, the triggers and how it reports — not just half of one. Handy for backups, for handing a client what they paid for, or for sharing a setup with the RAiN team.
When a run fails
| Message | What it means | What to do |
|---|---|---|
| "No … key connected" | An agent uses a provider without a stored key | Add your key, or switch that agent to Claude |
| "Daily run limit reached" | You've used today's quota | It resets at midnight UTC |
| A step shows an error mid-trace | That agent's work failed | Read the step, adjust the agent's goal/knowledge, run again |
| "Connection lost during the run" | Your network dropped while streaming | Check the Execution tab — the run itself usually completed |
FAQs
Do runs keep going if I close the tab? The run continues on RAiN's servers and its trace is saved — reopen the project's Execution tab to see the result.
Can I stop a run in progress? Not yet. Runs are short (seconds to a couple of minutes); a stop button is on the roadmap.
Why did my agent make something up? It likely lacked the facts. Put the truth in the project's Knowledge base (Settings) and tell the agent in its goal to answer only from the knowledge.
Where do I see my usage? Each run's trace is in the Execution tab. Workspace-level usage reporting for experts is coming; the RAiN team already monitors platform-wide usage.
A scheduled run didn't happen. Where do I look? The Execution tab: if it fired, there is a trace. If there is nothing, the schedule didn't fire — check the trigger is switched on and that its rule saved. See Trigger.
Do scheduled runs count against the daily limit? Yes. Every run counts the same, whoever or whatever started it.
Related: Quick start · Trigger · Endpoint · Designing agents in the studio · Models & LLM connections