Published July 19, 2026
CosmicDrift.io ships with two ways to play from the exact same menu: Play Solo and Launch (multiplayer). They look similar in the first ten seconds — same steering, same growth loop, same head-to-body death rule — and that similarity is deliberate. But they're not the same game wearing a different label. One runs entirely on your own machine against AI; the other puts you on a real server against real people, and the AI population shrinks specifically to make room for them. This post is about what's actually different, technically and in feel, so you can pick the one that matches what you're actually in the mood for on a given session.
Single-player CosmicDrift runs the entire simulation client-side — there's no server round-trip for anything that happens in your match. Your inputs, the bot AI, collision checks, orb spawns, the boss timer: all of it executes locally in your browser. The practical upshot is exactly what you'd expect from a local game with no network dependency: there's no latency to fight, no connection to drop, and nothing on a remote server that can desync from what you're seeing on screen. Every life is populated by a fixed 50 AI bots — not a number that scales with anything, just a constant the single-player mode always uses, since there's no server to offload bot AI or physics to and no real-player population to scale around in the first place.
Because those 50 bots are the entire population, single-player is a controlled environment in a real sense — the AI is well-tuned (wall avoidance, collision escape, orb targeting, zone gravity, perk accumulation all run for bots too), but it's still AI. It follows priority rules in a specific order, reacts to a specific lookahead distance, and behaves consistently enough that a player who studies it can learn its patterns. That's not a knock on the bots — a genuinely unpredictable bot population would just be worse AI, not better — it's just a structural fact worth being honest about: no amount of AI tuning erases the difference between "an opponent that reacts to state" and "an opponent that wants something and adapts."
Multiplayer connects you to a Colyseus WebSocket server, and the bot count in that room isn't fixed — it scales down from the same 50-bot ceiling toward 0 as real players fill the room. An empty or near-empty server looks a lot like single-player on the surface (mostly bots, same starting population), but the moment more real players join, the server makes room for them by pulling bots out rather than just piling humans on top of a static bot count. A busy multiplayer server, in other words, isn't "single-player plus some people" — it's a meaningfully different mix where more of what's on your screen is unpredictable human decision-making and less of it is AI.
That has a direct, practical consequence: collision, orb pickups, and every kill are server-authoritative in multiplayer. What the server decides happened is what happened, and your client is rendering and interpolating around that — which is also exactly why real network conditions matter in a way they structurally cannot in single-player. A great local connection makes multiplayer feel just as responsive as solo play; a rough one won't, and that's an honest tradeoff for playing against real people instead of a local simulation, not a bug to be fixed away entirely.
Here's the part worth stating plainly, because it's easy to assume single-player is a stripped-down "practice mode" version of the real game: it isn't. The 1,000-level stage system, all ten perks and their ten ranks each, the three drifting zones and their key-orb benefits, and the Cosmic Leviathan boss fight all exist in both modes, running off the same underlying formulas. That's not an accident or a coincidence of both modes reusing some old code — it's been an explicit, ongoing parity effort: past development passes on this project have specifically gone back and audited SP vs. MP behavior for mismatches (bot mass at spawn, body shapes, zone radii, orb counts) precisely because it's easy for two independently-running versions of the same systems to quietly drift apart, and worth catching when they do. If you learn the stage curve, a perk build, or how to fight the boss in single-player, that knowledge transfers directly — it's not a different game underneath the shared controls.
"No amount of AI tuning erases the difference between an opponent that reacts to state and an opponent that wants something and adapts — and that's fine. It's not a flaw in single-player, it's just what single-player is."
If you're new to the game, or you're trying something specific — a boss fight attempt, a particular perk build, figuring out how the collect-radius perk actually feels at rank V versus rank I — single-player is the honest place to do that. There's no risk of another real player converging on the same Leviathan spawn and taking the kill you were lining up, no risk of a stranger camping the zone you're trying to route toward. The 50-bot population is consistent enough that you can isolate one variable (your build, your positioning, your reaction time) and actually learn from the result instead of having it muddied by what a dozen unpredictable strangers decided to do at the same moment.
The thing single-player structurally cannot give you is genuine social tension — the specific feeling of a real person noticing you, deciding you're worth chasing, and adapting when your first evasive move doesn't work. Bots don't hold grudges, don't bait you deliberately, and don't get better at reading your habits over a match the way a real opponent can. Multiplayer is also the only place leaderboard placement means anything comparative — the home-screen leaderboard tracks best stage ever reached across real sessions, and a high stage earned by out-surviving and out-maneuvering actual people is a different kind of result than the same stage reached in an empty single-player room. If what you actually want is a real skill matchup with something at stake, that only exists in multiplayer.
Because multiplayer's bot count scales toward the same 50-bot ceiling as the room empties, joining a quiet server gets you something close to the single-player feel with the small residual chance a real player shows up mid-session — not quite either mode cleanly, but a reasonable way to get low-pressure practice without fully leaving the live server. It's worth knowing this is possible rather than assuming multiplayer is always a full lobby of real threats.
| Aspect | Single Player | Multiplayer |
|---|---|---|
| Where it runs | Entirely client-side, no server round-trip | Colyseus WebSocket server, server-authoritative |
| Bot count | Fixed at 50, always | Scales from 50 down toward 0 as real players join |
| Network dependency | None — no lag, no dropped connection | Real, since collision/kills are server-decided |
| Opponent behavior | Consistent AI, learnable patterns | Real players — adaptive and genuinely unpredictable |
| Stage/perk/zone/boss systems | Same systems, same formulas | Same systems, same formulas |
| Leaderboard bragging rights | Still saves your best stage, but against bots | The comparison that actually means something |
| Best for | Learning, practicing builds, boss attempts without competition | Real stakes, real skill matchups, genuine unpredictability |
Single-player and multiplayer in CosmicDrift.io aren't a "full game" and a "demo" — they're the same set of systems (stages, perks, zones, the boss) run in two genuinely different environments: one a fixed, local, 50-bot population with zero network dependency, the other a live server where the bot count actively makes room for however many real players show up. Neither is the objectively correct choice. Single-player is the honest, lower-pressure place to learn a build or practice a boss fight without a stranger taking the kill out from under you. Multiplayer is the only place the game's core tension — another mind on the other end of that trail, actually trying to beat you — is real rather than simulated. Most players will probably end up using both, for different reasons, in the same week.