What is upstreamapi.com?
Upstream is a rollout intelligence platform that goes beyond classic feature flags by not only determining whether a feature is on, but also whether it should stay on. It uses AI to monitor error rate, latency, and conversion metrics in real time. When metrics drift, Upstream automatically acts—ramping up gradually if healthy, pausing or rolling back if anomalies are detected. Rollbacks execute in a median of 0.2 seconds, safeguarding thousands of users before a human even sees an alert.
The platform offers dynamic config via a REST API with edge latency under 10ms, a Codebase Intelligence feature that maps flag usage and auto-generates documentation, and a Config Copilot that allows plain-English queries. Upstream integrates with Slack, Linear, GitHub, and supports migration from LaunchDarkly, Split.io, and Statsig. Pricing starts with a free tier and scales to enterprise with dedicated infrastructure and SLA guarantees.
The platform offers dynamic config via a REST API with edge latency under 10ms, a Codebase Intelligence feature that maps flag usage and auto-generates documentation, and a Config Copilot that allows plain-English queries. Upstream integrates with Slack, Linear, GitHub, and supports migration from LaunchDarkly, Split.io, and Statsig. Pricing starts with a free tier and scales to enterprise with dedicated infrastructure and SLA guarantees.
Features
- AI Rollout Pilot: Define SLOs and Upstream automatically monitors metrics, ramp up percentages, or pauses rollouts when signals degrade.
- Anomaly Guard: Real-time anomaly detection on error rate, latency, and conversion. Auto-rollback or human-review pause based on severity.
- Codebase Intelligence: SDK introspects codebase to map every flag usage, detect dead flags (unused >30d), and auto-generate flag documentation.
- Config Copilot: Ask in plain English about flag status and get answers with 1-click actions.
- Dynamic Config API: Core REST API for flags, multi-env support, audience targeting, and percentage rollouts with edge latency <10ms.
- Auto-rollback: Median rollback time of 0.2 seconds, safeguarding thousands of users before a human reads the alert.
Use Cases
- Safe gradual rollouts of new features with automatic rollback on failure
- Monitoring and optimizing feature performance with real-time AI analysis
- Managing feature flags across multiple environments with no manual intervention
- Cleaning up dead flags and auto-generating documentation for existing flags
- Allowing non-technical team members to query flag status via natural language
FAQs
-
How does AI Rollout Pilot decide when to ramp up?
You define SLO targets (error rate, p95 latency, conversion). Upstream evaluates metrics over a rolling window and advances the rollout percentage only when all targets are met. If any signal degrades, it pauses and alerts. -
How fast is an auto-rollback?
Median rollback time is 0.2 seconds. Anomaly Guard runs at 1-second intervals, and the rollback API call is fully synchronous at the edge. Over 2,000 users are typically safeguarded before a human reads the alert. -
Does Upstream work with my existing feature flag setup?
Yes. We have migration importers for LaunchDarkly, Split.io, and Statsig JSON exports. Flags come over with their targeting rules intact. You can run both in parallel during transition. -
What happens if Upstream goes down during a rollout?
Our SDK evaluates flags locally with a cached snapshot. If the edge cannot be reached for over 30 seconds, flags default to the last known stable state. We maintain 99.98% uptime with a committed SLA on Scale and above. -
What counts as an evaluation?
An evaluation is a single SDK call that checks the state of one flag for one user context. Batch evaluations (checking multiple flags at once) count as one evaluation per flag in the batch.
Helpful for people in the following professions
upstreamapi.com Uptime Monitor
Average Uptime
0%
Average Response Time
0 ms
