Ship AI changes
without the risk.
Canary deployments for LLM prompts and models. Repath splits traffic, scores every response with an AI judge, and rolls back automatically before your users notice.
The Repath Platform
Catch regressions, evaluate quality, and revert automatically — all before your users notice.
Mean responses before a quality drop is flagged and rollback triggered.
Eval runs async — never touches your critical path. Your users never feel it.
How it worksTime from detection to full traffic restore on the stable version.
Ship prompt changes
without the risk.
Route a small % of traffic to a new prompt version. Quality gates control every step — traffic only advances when scores consistently hold.
QUALITY SCORE
0.87
was 0.91
THRESHOLD
0.85
configured
TRAFFIC
50%
to support-v2
STEP
3/4
advancing
Rollout Steps
score ≥ 0.85 for 10 minAI models break silently.
Zero HTTP errors. Zero exceptions. Just worse outputs your users silently abandon — and you don't know for days.
5% → 25% → 50% → 100%
Configurable quality gates at every step. Traffic only advances when scores hold.
Async LLM judge — zero latency overhead
Every response evaluated against your rubric. Results arrive in ~120ms, never blocking your users.
Auto-revert in under 500ms
Score drops below threshold? 100% traffic back to stable instantly. No on-call needed.
Silent provider switching
OpenAI down? We retry then silently failover to Anthropic or OpenRouter. Your app keeps running.
How it works
Three steps. Automatic.
No code changes beyond one line.
Split traffic
Point your app at Repath instead of OpenAI directly. We route a small % to your new prompt or model — users see nothing different.
Score every response
An LLM judge evaluates every response against your quality criteria. Async — never adds latency to your app.
Auto-advance or rollback
Quality holding? Traffic advances to 100%. Quality drops? Rollback in under 500ms, before your users notice.
Everything you need
Built specifically for AI deployment safety.
Not feature flags with an AI sticker.
Canary deployments
5% → 25% → 50% → 100% with configurable quality gates at each step.
LLM-as-judge evaluation
Define criteria in plain English. Scores every response asynchronously — never slows your app.
Auto-rollback in <500ms
Score drops below threshold? Back to baseline instantly. No manual intervention.
Provider failover
OpenAI down? We retry, then silently switch to Anthropic or OpenRouter. Your app keeps running.
Provider health tracking
Live error rates per provider. Know about outages before your users do.
Full audit trail
Every advance and rollback logged with exact scores. Complete visibility into every decision.
Simple pricing
7-day free trial on every plan.
No credit card required to start.
- 10,000 evals/mo
- 3 active rollouts
- OpenAI + Anthropic + Gemini
- Auto-rollback
- 7-day data retention
- Email alerts
- Dashboard + API
- 100,000 evals/mo
- Unlimited rollouts
- All providers + OpenRouter fallback
- Auto-rollback
- 90-day data retention
- Slack + webhook alerts
- Custom eval criteria
- Priority support
- Unlimited evals
- Dedicated infrastructure
- SSO / SAML
- Team RBAC
- 1-year data retention
- On-call support
- Custom SLA
All plans include provider failover, auto-rollback, and real-time dashboard. Cancel anytime.
Stop shipping AI blind.
Know if your prompt change is better or worse — before your users do. Start your free trial in 30 seconds.
Start free trial — no card needed