A communication protocol where agents exchange spectral structure instead of text or JSON. The Laplacian IS the message. The conservation ratio IS the confidence. The Fiedler vector IS the routing.
- SpectralMessage: a message IS a Laplacian — encode agent state as a spectral object
- Spectral communication bus: agents register with capability graphs, send tasks as spectral projections, receive results as Laplacians
- Conservation-based confidence: the conservation ratio of the Laplacian tells you how well-structured the agent's response is
- Fiedler-based routing: tasks are routed to agents based on eigenvector alignment, not string matching
- No serialization overhead: pure NumPy linear algebra replaces JSON/text parsing
pip install numpy
python demo.pyThis runs 4 agents communicating through spectral messages — registering capabilities, sending tasks, receiving results, all without a single JSON object.
spectral_message.py — SpectralMessage: a message IS a Laplacian
comm_bus.py — AgentNativeComm: the spectral communication bus
demo.py — 4 agents communicating purely through spectral messages
| Concept | What it replaces |
|---|---|
| Laplacian | JSON payload — encodes full capability structure |
| Eigenvalues | Type/schema — spectral fingerprint identifies the agent |
| Conservation ratio | Status code — confidence in the message |
| Fiedler vector | Routing table — which agents should handle this |
| Spectral alignment | Content negotiation — cosine similarity of spectra |
# Traditional:
Agent A → "Please analyze this data" → Agent B → {"result": 0.95} → Agent A
# Agent-native:
Agent A → [Laplacian L_A + task vector t] → Agent B → [Laplacian L_B + result vector r] → Agent A
Part of the SuperInstance OpenConstruct ecosystem. This is the communication layer for:
- agent-spectrum-os — the OS that schedules agents spectrally
- agent-manifest-rs — manifests define the capability graphs that become Laplacians
- agent-handshake-rs — traditional handshake can upgrade to spectral communication
Integration tests via demo.py — 4 agents exchange spectral messages and assert successful communication, conservation ratios, and eigenvalue alignment.
pip install numpyPython 3 with NumPy. No other dependencies.