aacp-langchain 0.1.0

AACP coordination layer for LangChain multi-agent workflows

aacp-langchain

AACP coordination layer for LangChain multi-agent workflows.

AACP replaces natural language agent-to-agent instructions with typed, validated, deterministic packets. LangChain handles agent lifecycle, memory, and tooling. AACP handles what the agents say to each other.

Install

pip install aacp-langchain

Quick start

from aacp_langchain import AACPOrchestrator

orch = AACPOrchestrator(model="gpt-4.1-mini")
result = orch.run_workflow("payroll", period="2026-03")
print(result.summary())

Measured results

Payroll workflow comparison. gpt-4.1-mini. June 2026.

                        WITHOUT AACP    WITH AACP
Coordination LLM calls:        4            0
Coordination cost:          $0.0024      $0.0000
Agent cost:                 $0.0035      $0.0035
Total cost:                 $0.0059      $0.0035
Total saving:                             18%
Coordination deterministic:    NO          YES
Schema validated:              NO          YES

Department day comparison (59 coordination hops)

python3 examples/department_comparison.py --mock

                        WITHOUT AACP    WITH AACP
Coordination LLM calls:       59            0
Coordination cost:         $0.0031      $0.0000
Total saving:                             18%
Deterministic:                 NO          YES

Without AACP vs With AACP

Without aacp-langchain:
  Orchestrator
    ↓ "Please ask the HR agent to retrieve salary records..."
  LLM call  ← costs tokens, varies every run
    ↓ "HR Agent, could you fetch the employee salary data?"
  HR Agent

With aacp-langchain:
  Orchestrator
    ↓ FETCH|HR|return:ORCHESTRATOR|p:1|aacp:1.1|res:emp_salary|period:2026-03
  HR Agent  ✓ validates. $0.00 encoding.

Workflows

# Payroll (5 hops)
result = orch.run_workflow("payroll", period="2026-03")

# IT provisioning / JML onboarding (6 hops)
result = orch.run_workflow(
    "it_provisioning",
    username="j.smith",
    dept="Engineering",
    licences=["M365", "Slack"],
)

# Sales qualification (5 hops)
result = orch.run_workflow(
    "sales_qualification",
    lead_id="L-001",
    lead={"budget_gbp": 75000, "timeline_months": 3},
)

Requirements

• Python 3.10+
• OPENAI_API_KEY
• pip install aacp-langchain

Links

• Protocol spec: https://aacp.dev
• Python SDK: https://github.com/MackayAndrew/aacp
• CrewAI integration: https://github.com/MackayAndrew/aacp-crewai
• Community rules (241): https://github.com/MackayAndrew/aacp-community-rules
• IETF Draft: https://datatracker.ietf.org/doc/draft-mackay-aacp/

Licence

MIT