Saykai

Deterministic safety gate for robots and high-stakes agents.

On every pull request, Saykai reruns your scenarios and log replays, checks them against your Safety Spec (your versioned definition of safe enough), and blocks changes that fall below your safety bar.

For teams running agentic systems, physical AI, and other high-stakes software that cannot afford regressions.

Example output: 10/10 scenarios passed. Safety Pack generated. Blocks only when your Safety Spec fails.
CI_PIPELINE // PR-829
Commit
PASSED
Deploy
Status: PASSED Time: 14s

Safety Spec = versioned rules  •  Safety Pack = JSON evidence per run

Teams ship faster than they can validate real-world behavior changes.

Saykai turns CI into the first safety line before anything reaches production hardware, money, or users.

How it fits into your pipeline.

Commit → Saykai Gate → Deploy
Runs as a required check on every pull request.

Drop Saykai into your existing CI as a required check.

  • Every PR runs safety scenarios and log replays
  • Results are compared to your Safety Spec and baseline
  • A Safety Pack explains the decision and what changed
  • The PR passes or blocks based on policy, not gut feel

Runs in CI. You control what data is included in scenarios and log replays.

> Drop in Saykai once, then every change is screened against the same safety bar.

Saykai Gate
Required • Passed in 14s
10/10 scenarios passed
Latency regression: +0.2% (allowed)
Safety Pack generated: sp-8a7b9c.json
Example gate output shown above.

Start with one CI integration.

Step 1

Define your Safety Spec (saykai.yml)

Describe what safe enough means for one system. Scenarios, metrics, thresholds, and blocking rules live in version control with your code.

Step 2

Install the GitHub Action

Run Saykai as a required check on every pull request in one critical repo.

If you are not on GitHub Actions, we will help you wire it in during the pilot.

Step 3

Get a Safety Pack on every PR

Each run produces a JSON Safety Pack with metrics, diffs vs baseline, and a clear pass or block decision you can hand to safety and risk.

Private beta. Accepting a small number of pilot teams in industrial robotics and high-risk agent workflows.

Pilot with Saykai

  • Define your Safety Spec: Work together to classify a critical workflow and write the first saykai.yml.
  • Integrate the CI gate: Wire the gate into a single pipeline, starting with GitHub Actions.
  • See value fast: Start receiving Safety Packs on every PR within 2 weeks.
No lock-in. If it is not catching useful issues in the first month, you stop.
GUARANTEE.TXT

NO_LONG_TERM_CONTRACT

IF VALUE == FALSE:
  EXIT(0)

No lock-in, no long-term contract.

Gate your next deployment.

The first step is simple: add saykai.yml and the Saykai CI check to one critical pipeline. From there, every change to that system is screened through the same deterministic safety bar.

Request pilot access View GitHub reference examples
Security Policy