Reduce your ping.
Free. Forever.

There is no catch. The community relay network is community-hosted and sponsor-funded, so you can install the client and play without running anything yourself. The software is fully open source, and you can still self-host your own proxy if you prefer: Vultr gives new accounts $300 free credit and Oracle Cloud has a free tier forever. We don't sell your data. Anonymous aggregate telemetry is on by default and can be turned off any time; see the Privacy Policy.

LightSpeed is designed to be compatible with major anti-cheat systems: it does not modify game files or memory, does not hook game processes, and does not hide your IP. Route optimizers in this class have been used for years. That said, anti-cheat policy and terms of service are set by each publisher and can change at any time, so no tool can guarantee a ban will never occur. Check the relevant game's terms before you play.

It depends on your location and ISP. If your ISP routes your traffic through suboptimal paths (very common), LightSpeed can cut a meaningful amount off your ping. If your ISP already has optimal routing, the improvement may be minimal. The model picks the best route, or direct if that's fastest. The hero stat above is an estimate from real client measurements: it is the difference between the client's ICMP round trip to the game server and its tunnelled game-traffic round trip. ICMP and game traffic are different instruments, so read it as a signal, not a promise of in-game latency.

No. LightSpeed can't make your traffic go faster. It can't outrun physics. What it does is proactively route your traffic onto the fastest available path, avoiding congestion and needlessly long routes. If your ISP already has a clean route, or your proxy is badly placed, you'll see little to no improvement.

Usually one of two things: (1) the proxy is farther from the game server than your direct path, so the extra hop adds latency, or (2) the proxy's data center isn't well-connected to the internet backbone. Proxy placement matters a lot: how close the datacenter sits to major backbone/peering points. Real-world routing is also governed by BGP peering contracts and ISP prioritization, which neither you nor LightSpeed control. Pick the proxy closest to the game server, not closest to you. For an interactive look at how routing works, see the Route Visualizer.

Currently 24 games: Fortnite, CS2, CS:GO Legacy, Dota 2, Rust, Valorant, Apex Legends, Overwatch 2, League of Legends, PUBG, MapleStory, Genshin Impact, Rocket League, World of Tanks, Dead by Daylight, Bodycam, Roblox, Project Zomboid, WARDOGS, Minecraft, Hunt: Showdown, Rainbow Six Siege, and ARC Raiders all have built-in profiles. Just run with --game <name> or let auto-detection pick it up. Any other UDP-based game can work with manual port configuration.

Anonymous aggregate telemetry is on by default. No accounts, no registration, and no personal data. It sends only aggregate network metrics (latency percentiles, jitter, FEC counters) over plaintext HTTP to the relay's unauthenticated /telemetry endpoint: no IP addresses, no identifiers, and no packet contents. A cell is suppressed until it has at least 3 reports (this counts reports, not distinct people). Turn it off any time with --no-telemetry, telemetry = false in lightspeed.toml, or the GUI checkbox. All traffic passes through unencrypted to avoid adding overhead. See the Privacy Policy and Data Dictionary.

The easiest path is to do nothing: the default client experience uses the community relay network, and your client auto-discovers the fastest relay from the signed registry. If you want to run your own node instead, spin up a $5-6/mo VPS on Vultr (or use Oracle Cloud free tier), run the setup-new-node.sh script in the repo, and point your client at your own IP. It takes about 5 minutes. Data, control, and health ports are configurable via the [network] section of the proxy config. See infra/README.md for the full guide.

Rust gives us the performance of C/C++ with memory safety guarantees. For a latency-critical network tool that handles thousands of packets per second, we need zero-overhead abstractions. Rust's async ecosystem (Tokio) and QUIC libraries (quinn) are world-class.

We train a Random Forest model on network latency data with 11 features: current latency, historical p50/p95, jitter, packet loss, hop count, time of day, proxy load, and geographic distance. The model predicts which proxy route will be fastest in a single inference pass, measured in microseconds at runtime. The model reports its own R², MAE, and RMSE against a held-out test split each time it trains; the figure varies with the training data, so we do not publish a fixed number.