Skip to main content
Runtime is where operations teams create and run AI agents. A payments support team gives an agent the job of reconstructing a disputed charge, a risk team gives one the job of assembling a fraud review packet, an onboarding team gives one the job of preparing an underwriting memo. Each agent has a defined job, a measurable success criterion, scoped access to the team’s tools, a runbook for how it works, and an approval gate in front of anything consequential. Agents run inside isolated cloud sandboxes, so a mistake stays contained, and every run is recorded end to end: the prompt, the tool calls, the cost, the result, and the grade.

​
What an agent is made of

Every agent in Runtime is built in six steps, and the order matters: make it capable first, prove it on real cases with read-only access, then lock it down. The dashboard, the CLI and these docs all follow the same order.

Quickstart: build your first agent

Ten minutes from a blank org to a graded run.

Build an agent

The six steps, one page each, with the dashboard steps and the CLI behind them.

Guides

Worked agents for payments operations: support, fraud and risk review, merchant underwriting.

API Reference

Every endpoint, plus the runtm-api CLI that wraps them.

​
How a run happens

  1. A trigger fires: someone mentions the agent in Slack, an email lands in its inbox, a Linear issue is assigned to it, or a cron tick arrives.
  2. Runtime launches a session on the agent’s default template, with the agent’s instructions layered on top of the org’s and the template’s.
  3. The agent works the case using the tools and skills the template carries. Once guardrails are in place, anything they mark as sensitive either prompts, is denied, or waits for a human approval.
  4. The result goes back where the trigger came from: a thread reply, an email, a Linear comment, a PR.
  5. The grader scores the run against the agent’s evaluation categories, and the scorecard updates.

​
Hosted or self-hosted

The hosted platform at app.runtm.com. Agents, templates, triggers, guardrails, grading and the scorecard all live here. This is what the Documentation, Guides and API Reference tabs describe.

​
Programmatic access

Everything the dashboard does is also an API call. Create an API key in Settings > API Keys, then call the Cloud API at:
The runtm-api CLI wraps the whole surface and is what coding agents use when they build or operate other agents. Install it with curl -fsSL https://runtm.com/install | bash, set RUNTM_API_KEY, and see the Agent CLI reference. Throughout Build and Guides, every dashboard step has its CLI equivalent alongside it.