The real-time event layer for AI agents on Solana. Subscribe to any event in plain English. Get pushed the instant it matches.
plays nicely with the entire stack
THE PROBLEM
Every minute, agents fire off LLM calls just to ask the world "did anything happen?" It's slow, it's expensive, and it burns compute on silence.
polling agent · 1h window
50 LLM calls
2
real signals
48
pure burn
one purple cell = a real event · the rest? compute spent for silence.
THE SOLUTION
Open Source
eventmesh-engine — the core is yours →
E2E DEMO · COMING SOON
A 60-second walkthrough showing how eventmesh fuses every source into one push channel — and how your agent reacts the moment it fires.
01
subscribe
Subscribe to any event in plain English. No API keys. No custom connectors.
02
listen
Eventmesh fuses every source — onchain, prices, news, social — into one stream. Your agent doesn't poll. It listens.
03
act
The instant a condition matches, your agent gets pushed. Sub-15s latency. Pay per event via x402 on Solana.
BENCHMARK
We ran two agents on the same watchlist. The polling agent woke up 785 times to fetch — only 120 of those had anything new to react to. The other 665 were pure burn. Eventmesh just pushes the 120 events that mattered. Drag the slider to see how it scales.
Matched events
120
6.5×
fewer calls
5.4×
cheaper
3.2×
fewer tokens
Per-call polling
$20.33
785 cron wake-ups · 665 returned nothing.
With Eventmesh
$3.76· $16.57 saved
120 pushes · pay only when an event actually fires.
THE INTERFACE
Every action — login, topup, subscribe, list, cancel — runs through the same CLI. The agent reads its own state, debugs its own subs, and stays fully in control of its event stream — no human in the loop required.
AGENT INTEGRATION · ONE STEP
Point any agent at eventmesh.cc/skill.md — the agent learns to subscribe, listen, and react in one step.
Works with the agents you already use
POWERED BY EVENTMESH
READY?
One command installs the skill. Free during beta. No keys to provision.
eventmesh.cc/skill.md
Join the crew