Pre-alpha · Codex Labs LLC
The OS that
knows itself.
Zeos is the first operating system with proprioception. It measures its own silicon. It learns its own hardware. It resolves signal graphs instead of executing instructions.
Z+ is its native language. → is the only operator you need.
Drag to see the difference.
Same program. Same features. Thousands of times less code.
Microservices · Java/Go/Python · ~2M lines
16,667x less code
Z+ · 120 lines
// Timeline Service (Go) — 1 of ~200 microservices
package timeline
import (
"context"
"github.com/twitter/timeline/cache"
"github.com/twitter/timeline/fanout"
"github.com/twitter/timeline/storage"
"github.com/twitter/timeline/pubsub"
)
type TimelineService struct {
cache *cache.RedisCluster
store *storage.Manhattan
fanout *fanout.Service
pubsub *pubsub.EventBus
metrics *metrics.Client
}
func (s *TimelineService) OnTweet(ctx context.Context, tweet *Tweet) error {
followers, err := s.store.GetFollowers(ctx, tweet.AuthorID)
if err != nil {
s.metrics.Incr("timeline.fanout.error")
return fmt.Errorf("get followers: %w", err)
}
for _, batch := range chunk(followers, 1000) {
if err := s.fanout.Push(ctx, batch, tweet); err != nil {
s.metrics.Incr("timeline.fanout.batch_error")
log.Error("fanout batch failed",
"author", tweet.AuthorID,
"batch_size", len(batch),
"error", err,
)
// retry logic, dead letter queue, etc.
s.pubsub.Publish("fanout.retry", RetryEvent{
Tweet: tweet,
Batch: batch,
Attempt: 1,
})
}
}
// Update author's own timeline
if err := s.cache.Prepend(ctx,
fmt.Sprintf("timeline:%s", tweet.AuthorID),
tweet.ID,
); err != nil {
return fmt.Errorf("cache update: %w", err)
}
s.metrics.Incr("timeline.fanout.success")
s.metrics.Histogram("timeline.fanout.followers",
float64(len(followers)))
return nil
}
// ... plus 199 more microservices for:
// auth, search, trending, notifications,
// media upload, content moderation, DMs,
// rate limiting, analytics, ads, ..."zp-comment">// chirp.zp — the whole thing user : auth.identity "zp-operator">-> { voice, "zp-comment">// what they say ears, "zp-comment">// who they follow pulse "zp-comment">// their activity } "zp-comment">// posting = emitting a signal voice "zp-operator">-> post : {text, media, timestamp: t} "zp-comment">// post goes three places at once post "zp-operator">-> { vault.store, ears.followers "zp-operator">~> timeline, content.index } "zp-comment">// follow is a wire. that's it. user_a.voice "zp-operator">-> user_b.ears "zp-comment">// timeline IS the follow graph ears "zp-operator">-> "zp-operator">| "zp-keyword">merge(by: timestamp) "zp-operator">| "zp-operator">-> timeline "zp-comment">// like = "zp-keyword">tap that amplifies post "zp-operator">~> like(user) "zp-operator">-> post.heat↑ "zp-comment">// trending = "zp-keyword">resonance content.index "zp-operator">-> extract(topics) "zp-operator">-> topic_stream topic_stream "zp-operator">-> "zp-keyword">resonance(threshold: "zp-number">1000, "zp-keyword">within: 1h) "zp-operator">-> trending "zp-comment">// moderation is structural post "zp-operator">-> safety "zp-operator">@ ai("toxic-detect.zdx") "zp-operator">-> "zp-keyword">gate(safe) "zp-operator">-> deliver "zp-comment">// the whole backend. "zp-number">120 lines. "zp-comment">// no microservices. no kafka. "zp-comment">// just wires.
→
Signal graphs, not instructions
Z+ programs describe connections, not procedures. The OS resolves them simultaneously. No loops. No threads. No locks. The wiring IS the program.
Δ
The OS feels its silicon
Every core, every card, every bus — measured continuously. Not spec sheets. Real-time telemetry. Your machine optimizes for its actual capability, not its SKU label.
~>
Everything is the same pattern
27 programs across every industry. Healthcare, finance, robotics, gaming. All the same architecture: signal in → filter → route → signal out
27
programs built
2,458
total lines of Z+
38M+
conventional LOC replaced
15,000x
average compression
The entire language fits on a card.
Flow
→ signal flows
~> tap (observe only)
-x> sever (disconnect)
<-> exchange (request/response)
@ pin to hardware
Structure
| merge (converge)
{} fork (split)
Logic
gate pass or block
knee smooth transition
delta rate of change
resonance convergent evidence
Time
t t-1 t-2 temporal access
on_silence absence is data
baseline learned normal
sustained must hold for duration
Two tracks. Same language.
Game code IS robot code. The curtain lifts, the language doesn't change.