A communication-first architecture for agent systems.
`bca2p` separates the meaning of communication from the mechanics of execution. Signals are typed, routed, checkpointed, attributed, and adapted through a layered system.
Protocol layer
Defines the stable language of communication: signal envelopes, receptors, trust levels, artifact references, complexes, quorum rules, and topology policies.
Graph and runtime layer
Executes communication in super-steps, applies channel updates, checkpoints state, and keeps signal visibility deterministic enough for replay.
Learning and causal layer
Records signal lineage, ingests causal feedback, scores route contribution, and supports counterfactual questions about topology and amplification.
Transport and distributed layer
Moves signals across process and system boundaries while preserving payload structure, artifact references, and trace metadata.
Experimental research layer
Explores communication-policy training, biology-faithful simulation, native runtime replacement, and distributed substrate ownership.
Protocol layer
Signals, receptors, trust levels, complexes, quorum rules, topology policies, and artifact references define what communication means.
Runtime layer
Graphs, channels, super-steps, checkpointing, replay, and native execution define how communication moves through a workflow.
Learning layer
Causal feedback, contribution scoring, and counterfactual analysis define how the system adapts without losing traceability.
Network layer
Registry, transport, A2A bridging, mesh routing, and artifact movement define how communication crosses process boundaries.
| Stable platform path | Experimental research path |
|---|---|
| Typed signal envelope and receptor matching | Differentiable communication-policy training |
| Checkpointed runtime and replay | Biology-faithful cell signaling simulator |
| Registry, transport, and A2A bridge | SDK-owned distributed mesh substrate |
| Observability and causal feedback | Native runtime replacement path |
The implementation path keeps the usable SDK stable while the research platform expands around it.
Stable communication core
Typed signals, receptors, channels, checkpointing, replay, and graph authoring for practical agent systems.
Learning and observability
Causal feedback, counterfactual analysis, diagnostics, and transport-level traceability.
Adapter and native execution paths
Existing framework adapters alongside a native runtime for teams that want tighter ownership of coordination semantics.
Research extensions
Communication policy training, biology-faithful simulation, and distributed substrate experiments.