The SignalFlag platform is now open to try for free. Sign up, connect your agent, and run your first test. You get $100 in compute credit, and with no credit card requirement, you can be running tests in minutes.
Our new free tier features the full SignalFlag platform: MCP, web app, SDK, and CLI, plus the hardware-in-the-loop agent, queryable metrics, dashboards, CI integration, and SSO. Seats are unlimited on every SignalFlag plan, including the free tier.
Why we did this
SignalFlag’s evals platform is the result of years of hard work in collaboration with our amazing partners. From the outset one of the big challenges we faced was integration. Every customer has a unique robot, stack and workflow. This made white-glove support from us essential to a successful integration. We have taken these learnings, refined our processes and docs and - most importantly - built these integration learnings into our agent-skill via our MCP. Now, we’re extremely proud to open up our platform to anyone and empower them to self integrate using our MCP.
Run your first test on your own build
The first thing most developers want to do is select their branch and run it. Your agent can wire that up: Connect your agent to our MCP, and run your first batch. Or you can do it yourself: install the CLI, and point SignalFlag at your build. About 10 minutes, no Docker required.
Results come back in one place: pass and fail rates across the batch, metrics you can sort and filter, charts you can compare against the last build. When a number looks wrong, click into it and find the run behind it.
Passing tells you nothing broke. Comparing your metrics across builds tells you whether the robot is truly improving.
Give your agent a running start
Your coding agent can get a test harness running quickly. Building one you can rely on takes considerably longer. It becomes code your team owns, debugs, and maintains. It will also be built for scale as usage grows across more tests, more runs, and more people.
So the question isn't whether your agent can build it. It's what it builds on.
Connect your agent to our MCP, and it builds against components already running in production instead of writing that foundation from scratch.
Same agent, same prompt, less code to own. And because everyone on your team is building against the same components, what one person sets up is legible to the next.
If you're not sure what to measure
What you measure is the part most teams have yet to document: whether the planner is choosing better paths than last month, whether detection quality holds at dusk, what the distribution looks like rather than the single run.
So don't start from an empty project. Bootstrap a demo project, and you can see how one is put together end to end, with metrics already defined and results already flowing. Work through the tutorials if you'd rather build it up piece by piece. And when you're ready to define your own, metrics templates give you a working structure to change rather than a blank file to fill.
The docs are also published as plain markdown at llms.txt, so you can point your agent at them and let it read the whole thing before it writes anything, whether or not you're using the SignalFlag MCP.
What comes next
Seats are unlimited with the free tier, and nothing you build gets left behind if your usage grows. Compute does scale with usage, so once your team needs dedicated compute, deployment team support, or help building out test scenarios, get in touch.
Every team in physical AI ends up building similar test infrastructure. Yours doesn't have to. Start for free.