About Signa
Risk people, building for models that never sleep.
Founded in London in 2023 by people who ran model risk at two of Europe’s largest banks. Today 64 engineers and researchers build the evaluation layer regulated teams use to ship AI agents they can defend.
01 // Story
We spent years signing off models we could not fully see. So we built the tool we wanted: one place to test, trace and guard every agent — and prove it.
02 // Principles
How we work.
01
One loop, not three tools
Evals, tracing and guardrails share one data model — so a failure in production becomes a test case by the next deploy.
02
Measured, not guessed
Every metric we ship is validated against human-graded sets first. If a score can’t be explained to an auditor, it doesn’t ship.
03
Your data stays yours
Prompts, traces and customer data stay in your cloud. We never train on it, and you can delete it in one call.
03 // Milestones
Three years. Thirty-eight teams.
Q1 23
Founded in London; first eval suite ships to a tier-one bank.
Q4 23
New York office opens for US insurers.
Q2 24
Tracing launches; 100M spans captured in month one.
Q1 25
First EU AI Act conformity file built on Signa.
Q3 25
Guardrails go live in under 40 ms at the p95.
Q2 26
4.2 billion traces evaluated to date.
Ship the agent. Keep the evidence.
Send us an agent, a failure or a policy. An evaluation engineer replies within one working day.