# RunAgain pricing > Start free. Pay when it runs in production. Every plan is the whole loop: trace, experiment, test, evaluate, improve, monitor. The tiers differ in how much you send us, how long we keep it, and how far the loop closes on its own. ## Free The whole loop, on your first agent. - 1,000 events / mo - 14-day data retention - Alerts for failures - Experiments - Self-improving agents loop - 50 evaluations ## Pro — $399/mo Production agents that get better every week. - Everything in Free - 1,000,000 events / mo - 6-month data retention - Shared Slack channel - Self-improving agents loop - Unlimited evaluations ## Custom Security, controls and scale on your terms. - Everything in Pro - Custom event volume & retention - SSO - Audit log - Custom evals - Custom SLAs ## In every plan - Tracing for every run - Replay in mocked environments - Trace search and diffs ## FAQ ### Is my data secure? The honest answer: we process exactly the traces and conversation data you choose to send us, and nothing else. Send pseudonymous IDs and redact secrets and sensitive fields before ingestion, in your app or in the SDK. Transport is HTTPS, and dashboard access is authenticated. We do not claim to automatically strip PII for you, because a redactor that silently misses a field is worse than none. If you need a security review, a DPA or a specific deployment setup, talk to us and we will do it properly. ### Can we sign a DPA? Yes. On your traces we are the processor and you are the controller, so a data processing agreement is the normal way to run this. Email tamas@runagain.ai: we will send ours, or review and sign yours. We will name our sub-processors, confirm where data sits and how long we keep it for your tier, and delete your data on request. Custom adds the rest of what procurement usually asks for: SSO, audit logs, retention you set, and SLAs in writing. ### How complicated is the setup? You do not rebuild your agent and you do not change how it works. Point RunAgain at the runs you already produce: wrap your entrypoint with the SDK, or send OpenTelemetry spans you are already emitting. Watch one staging trace land, confirm you are happy with what is on it, then turn it on in production. First trace in minutes, not a quarter-long integration project. ### What counts as an event? One event is one step recorded inside a run: a model call, a tool call, a retrieval, a span you emit yourself. A short agent turn is a handful of events; a long autonomous run can be dozens. Evaluations are counted separately, so scoring your traces never eats into your event allowance. ### What happens if I go over the limit? We keep ingesting and tell you, rather than silently dropping the traces you will need most on the day something breaks. You will see it in the dashboard and hear about it before it turns into a bill surprise. If your volume has genuinely outgrown the tier, we resize it with you. ### Does it work with the framework I already use? Yes, if it can emit a span. There is zero-config capture for the Vercel AI SDK and the Claude Agent SDK, an MCP server, plain HTTP, and OpenTelemetry ingestion that covers the OpenAI Agents SDK, LangChain, LangGraph, LlamaIndex, CrewAI, AutoGen, Mastra, Pydantic AI, DSPy, Google ADK and the rest. If you have a homegrown loop, the REST API takes it. ### Does replaying a run hit my production tools and paid APIs? No. Replays execute against recorded tool responses from previous runs, so a replay has no side effects: no emails sent, no rows written, no per-call charges from the APIs your agent talks to. That is what makes iterating fast and free, and what makes two versions of an agent comparable at all. ### Can I bring my own evals and judges? Yes. Use the built-in catalog of scorers, rubrics, assertions and trajectory checks, write your own judge prompts against your own criteria, or push scores in from an eval harness you already trust. Free includes 50 evaluations a month, Pro is unlimited, and Custom covers evals we build with you. ### What does the self-improving loop actually do? It closes the circle instead of just reporting on it. Failed evals, tool errors and drift become drafted tests and suggested fixes, delivered to your shared Slack channel with the traces that justify them. You review and approve; nothing ships to your agent on its own. Then it measures whether the change moved your goals and conversions, or only looked like it did. Get a demo: https://cal.com/tamas-szuromi/30min · Get in touch: tamas@runagain.ai --- Markdown mirror of https://runagain.ai/pricing for agents and LLMs. Append .md to any runagain.ai page URL for its markdown twin. Overview: https://runagain.ai/llms.txt · For agents: https://runagain.ai/for-agents.md