DEV Community

Mavos.by.Kyklos
Mavos.by.Kyklos

Posted on

SignalMesh: The Open Source Ambient Context Layer for AI Agent Fleets

#opensource #llm #agents #ai

99.97% cost reduction on context reads. 1.69µs retrieval. Drop-in with LangChain, CrewAI, AutoGen.

The problem every multi-agent system has

Your agents are making tool calls to read context that hasn't changed. Each one costs:

  • 800ms+ round-trip latency
  • Scaffold tokens burned on the same boilerplate
  • API cost, repeated per agent, per request

With 5 agents and 3 context reads each: $1,387/year on reads alone.

SignalMesh: broadcast once, tune in everywhere 

pip install signalmesh  # or self-host via Docker
Enter fullscreen mode Exit fullscreen mode 
from signalmesh import signal_registry

# Any source broadcasts
signal_registry.broadcast("market_data", "rss", {"btc": 42000})

# Any agent tunes in — 1.69µs, no network, no tokens
context = signal_registry.tune_in(["market_data", "price"])
Enter fullscreen mode Exit fullscreen mode

The mesh is in-memory, per-frequency buffered (last 100 signals), and keyword-flexible — agents find context even when their keyword doesn't exactly match the frequency name.

What's live right now

The public mesh is running at https://acecalisto3-signalmesh.hf.space:

  • 27 active frequencies
  • Real external agent traffic
  • CORS open, no auth required
  • 7 REST endpoints 
curl https://acecalisto3-signalmesh.hf.space/ui/frequencies      # all live frequencies
curl https://acecalisto3-signalmesh.hf.space/ui/status           # mesh health + signal count
Enter fullscreen mode Exit fullscreen mode

The numbers

Metric Value
tune_in() latency (single agent) 1.69 µs
tune_in() latency (100 concurrent) ~1.25 ms
Cost vs tool call architecture -99.97%
Payload size impact on latency negligible (refs, not copies)

Works with your existing stack

No schema changes. No migration. Broadcast from wherever you produce context:

# LangChain tool → mesh
@tool
def fetch_and_broadcast(query: str):
    data = your_api.get(query)
    signal_registry.broadcast(query, "tool", data)
    return data

# CrewAI agent reads from mesh instead of calling tool
context = signal_registry.tune_in(["query_keyword"])
Enter fullscreen mode Exit fullscreen mode

Tiers

Open Source Managed Cloud Enterprise
Price Free (MIT) $299/mo Custom
Nodes Unlimited (self-host) 500 Unlimited
SLA — 99.9% 99.99%
Support Community Email + Slack Dedicated engineer

Custom implementations (LangGraph, AutoGen, CrewAI integration) available — flat-rate, delivery in days.

Links

Top comments (0)