SignalMesh: Ambient Context for Agentic Fleets

Most multi-agent systems suffer from the same problem: agents that can't hear each other without being explicitly wired together. SignalMesh fixes this with one primitive — the signal.

What is a signal?

A signal is a named, timestamped broadcast. Any agent can emit one. Any agent tuned to that frequency receives it — no polling, no point-to-point plumbing, no orchestrator in the middle.

# Agent A emits
signal_registry.broadcast(name="market.btc.price", data={"price": 67420})

# Agent B receives — no direct connection to A required
@on_signal("market.btc.price") 
def handle(signal):
    print(signal.data)  # {"price": 67420}

No tool calls. No shared memory. No bottleneck.

Traditional frameworks route everything through a central coordinator. SignalMesh is ambient — signals propagate through a frequency mesh. Agents tune in to what they care about. The rest is silence.

Why this matters for HuggingFace

Every Space, model, and dataset is a potential signal emitter. SignalMesh gives them a way to announce state changes to any listening agent — no custom integrations, no API tokens.

• A Space finishes a job → emits space.job.complete
• A dataset updates → emits dataset.new_rows
• An agent picks it up → acts

The AGENTS.md layer

Every SignalMesh node serves a machine-readable manifest at /agents.md — describing what frequencies it listens to, what it emits, and what it accepts.

curl https://acecalisto3-signalmesh.hf.space/agents.md

This is the agentic web, arriving.

Try it:

Hit the live Space

curl https://acecalisto3-signalmesh.hf.space/health

Broadcast a signal

curl -X POST https://acecalisto3-signalmesh.hf.space/broadcast \
  -H "Content-Type: application/json" \
  -d '{"name": "my.signal", "data": {"hello": "world"}}'

→ Space (https://huggingface.co/spaces/acecalisto3/SignalMesh)

→ GitHub (https://github.com/Ig0tU/SignalMesh)