About · v0.1-draft

A small, honest protocol.

What AACP is

AACP, Agent Action Compression Protocol, is a draft specification for a compact, bracketed key:value format used between agents in a multi-agent LLM system. It targets the coordination layer only: routing, references, hand-off metadata, structured intent.

The thesis is narrow: agent-to-agent prose is wasteful. Replacing it with a parseable, extensible token format reduces coordination cost by 85–91% on the workflows we measured, improves auditability, and gives a clear surface for validation.

What AACP is not

  • It is not a way to reduce the cost of the model doing the actual task work.
  • It is not a replacement for MCP, A2A, or function-calling specs, it's complementary.
  • It is not a binary compression scheme. It's plain text designed for humans to read.
  • It is not appropriate for emotional or relational context, those compress poorly.
  • It is not production-validated. v0.1 is a draft. Live benchmarks in progress.

Relationship to other work

ProtocolLayerRelationship
MCPTool / contextComplementary. AACP packets can be the payload between MCP-aware agents.
A2AAgent transportComplementary. AACP is the wire-format on top of A2A.
GibberlinkAcoustic / audioDifferent problem (transport medium). AACP is text-layer.

Prior work & related research

The problem AACP addresses is independently recognised in published research and practitioner documentation.

SourceFindingRelationship to AACP
EcoLANG
Mou et al., Fudan University, May 2025 · arXiv:2505.06904
"There exists redundancy in current agent communication: when expressing the same intention, agents tend to use lengthy and repetitive language." Achieved >20% token reduction through evolved compression language for social simulation.Same observed problem, different angle. EcoLANG compresses evolved natural language; AACP uses a structured packet schema and targets business workflow coordination, measuring 91% on coordination tokens.
Framework practitioners
LangGraph vs CrewAI vs AutoGen comparisons, 2025–2026
CrewAI's role-based prompts inflate token count by ~30–50% vs hand-tuned alternatives for equivalent tasks. AutoGen's multi-turn conversation model inflates further, every agent turn involves a full LLM call with accumulated conversation history.Confirms coordination-layer verbosity is observed across mainstream frameworks, not just in academic settings.

AACP's position · EcoLANG compresses evolved natural language. AACP uses a structured packet schema. Both address the same observed problem from different angles. Neither MCP nor A2A address the verbosity of message content, that is the gap AACP fills.

Roadmap

VersionFocus
v0.1 validatedMay 2026. First benchmark: payroll workflow, 91.4% coordination token reduction measured (Claude Sonnet 4.5, 3 runs).
v0.1-draftFormat, validator, demo, honest framing. Current.
v0.2Live API benchmarks across 3 model families. Published methodology.
v0.3Reference encoder/decoder libraries (TS + Python).
v0.4Reserved key registry. Domain key conventions.
v1.0Stable spec, conformance test suite, ecosystem proposals.

Open questions

  • How should multi-packet conversations reference prior packets? Hash? Sequence?
  • Is there a useful binary envelope for very high-throughput inter-agent buses?
  • What does a conformance test suite look like?
  • How do we encode partial / streaming intent without losing the structural advantage?

Issues and PRs welcome on GitHub.

Licence: MIT.