Upstream is an AI-native feature flag platform positioned as the control plane for secure software delivery. Its core idea is to upgrade feature flags from "whether to turn on" to "whether it should stay on": traditional flags only answer whether a feature is on, while Upstream also monitors metrics and, when signals such as error rate, latency, or conversion rate drift, takes action before alerts are triggered.
【Main Capabilities】
- Dynamic Configuration API: Provides a core REST API supporting feature flags, multiple environments, audience targeting, and percentage-based canary releases, with edge evaluation latency below 10ms.
- AI Rollout Pilot: After users define SLOs, Upstream automatically monitors metrics and gradually ramps up, or pauses the release when metrics are abnormal.
- Real-Time Anomaly Detection: Performs real-time anomaly detection on business and technical signals, and executes automatic rollback or manual review pause based on severity.
- Codebase Intelligence: The SDK can introspect the codebase, map where each flag is used, detect dead flags unused for more than 30 days, and automatically generate documentation.
- Config Copilot: Supports asking questions in natural language, such as "What are the active flags for Enterprise customers in the EU region?", and can directly execute one-click actions.
【How to Use】 The official description is three steps: 1) Connect the SDK and embed five lines of code in your codebase, in any language, and flags immediately begin evaluating at the edge; 2) Define SLOs and tell Upstream what healthy means, such as error rate, p95 latency, and conversion rate; 3) Release with confidence, and Upstream automatically ramps up, monitors, and rolls back, with users receiving Slack notifications instead of late-night alerts. The official description also mentions observable SLOs as the single source of truth, defined once and then automatically enforced continuously across all flags, releases, and environments.
【Applicable Scenarios】 Mainly aimed at SREs, platform engineering, and software teams that need safe releases, for feature canarying, risk control, and automated rollback. Official user cases mention migrating 80 flags from LaunchDarkly, AI Rollout Pilot replacing about 90% of manual ramp-up work, and Codebase Intelligence discovering 14 dead flags.
【Pricing and Limits】 Official transparent pricing is provided: the free tier is aimed at personal projects and experiments, including 10,000 evaluations per month, 7-day audit logs, and community support; the team tier supports up to 10 seats, including unlimited flags, 1 million evaluations per month, Slack/Linear/GitHub integrations, and 90-day audit logs; there is also an enterprise tier with unlimited seats, unlimited environments, unlimited evaluations, 1-year audit logs, dedicated infrastructure options, and priority support, and it offers SOC2, HIPAA, on-premises deployment, and custom SLO agreements. Specific price figures are not yet provided.
