plato-edge 0.1.0

Edge-optimized Cocapn fleet packages for ARM64 devices with limited resources

plato-edge

Edge-optimized Cocapn fleet packages for ARM64 devices with limited resources.

Target Hardware

• NVIDIA Jetson Orin (ARM64)
• 24 GB RAM shared with GPU
• Python 3.8+ (system default on JetPack)

Design Goals

Goal	Constraint
Zero external dependencies	stdlib only
Installed size	< 50 KB total
Runtime memory	< 10 MB RSS combined
Network	Works fully offline after install
ML frameworks	No numpy, scipy, torch, etc.

Modules

plato_edge.tile_spec

Lightweight tile format for spatial/data tiling.

• Compact binary encoding (~22-byte header + payload)
• Optional metadata as NUL-delimited key/value pairs
• Bounds validation with configurable limits
• No optional dependencies

from plato_edge import Tile, TileCodec

tile = Tile(x=1, y=2, z=3, payload=b"hello", meta={"k": 42})
data = TileCodec.encode(tile)
tile2 = TileCodec.decode(data)

plato_edge.deadband

P0/P1/P2 priority gate using minimal regex.

• No NLP, no heavy text processing
• Configurable regex patterns per priority level
• Fast default patterns for ALERT, WARN, HEARTBEAT, etc.

from plato_edge import classify, DeadbandGate

level = classify("ALERT: engine failure")  # -> DeadbandGate.P0

plato_edge.flywheel

In-memory pub/sub and key-value cache.

• Thread-safe with fine-grained locking
• Optional per-key TTL
• Automatic eviction when capacity limits are reached
• No persistence — everything is in-memory

from plato_edge import Flywheel

fw = Flywheel(max_kv=4096, max_queue=256)
fw.set("key", "value", ttl=60.0)
fw.subscribe("topic", lambda msg: print(msg))
fw.publish("topic", "hello")

plato_edge.keeper

UDP discovery beacon replacing HTTP-based service discovery.

• Zero-dependency UDP broadcast/multicast
• JSON ping/pong protocol
• Threaded listener with daemon thread

from plato_edge import Beacon

with Beacon(port=37201, identity={"role": "camera"}) as b:
    peers = b.ping(host="255.255.255.255", timeout=1.0)

plato_edge.explain

Stripped explainability — trace IDs only, no full audit log.

• 24-character hex trace IDs (time + random)
• In-memory span tracking with nanosecond resolution
• Automatic span eviction when limit is reached

from plato_edge import Tracer

tr = Tracer(max_spans=1024)
tid, t0 = tr.start("inference")
# ... work ...
tr.end(tid, t0, meta={"status": "ok"})
print(tr.spans(tid))

Installation

pip install plato-edge

Development

python -m pytest tests/

License

MIT