AFMX — Agent Flow Matrix Execution Engine

"Tag your agent's cognitive intent once. Get 60–90% cheaper LLM costs automatically, with a full audit trail. No rewrites. Works with LangGraph, CrewAI, OpenAI, MCP, and anything else you're already running."

The one-line pitch

from afmx import Node, CognitiveLayer

# Add ONE field to your existing node.
# AFMX auto-routes: cheap model for retrieval/action, premium for reasoning/planning.
node = Node(name="analyse", handler="my_agent", cognitive_layer=CognitiveLayer.REASON)
# → Claude Opus / GPT-4o selected automatically
# → cheaper layers get Haiku / gpt-4o-mini automatically
# → every node result logged to tamper-evident audit trail
# → visual heatmap shows cost + model per cell

AFMX is the production execution fabric for autonomous agents. Deterministic, fault-tolerant, built like infrastructure.

What is AFMX?

AFMX is a production-grade, deterministic execution fabric for autonomous agents. It is not an agent reasoning framework — it is the layer that controls how agents act reliably in production.

Your Agent Logic  (LangChain / LangGraph / CrewAI / OpenAI / custom Python)
        ↓
ExecutionMatrix   (DAG: nodes + edges + mode + abort policy)
        ↓
AFMXEngine
        ↓
Deterministic execution:
  retry · fallback · circuit breaker · hooks · events · audit · RBAC

Install

pip install afmx

With extras:

pip install "afmx[redis,metrics]"    # Redis store + Prometheus
pip install "afmx[full]"             # everything except framework adapters
pip install "afmx[dev]"              # development + testing toolchain

Quick Start

python3.10 -m afmx serve --reload
# API:       http://localhost:8100
# Docs:      http://localhost:8100/docs
# Dashboard: http://localhost:8100/afmx/ui

curl -s -X POST http://localhost:8100/afmx/execute \
  -H "Content-Type: application/json" \
  -d '{
    "matrix": {
      "name": "research-pipeline",
      "mode": "SEQUENTIAL",
      "nodes": [
        {"id":"analyst",  "name":"analyst",  "type":"AGENT","handler":"analyst_agent"},
        {"id":"writer",   "name":"writer",   "type":"AGENT","handler":"writer_agent"},
        {"id":"reviewer", "name":"reviewer", "type":"AGENT","handler":"reviewer_agent"}
      ],
      "edges": [
        {"from":"analyst","to":"writer"},
        {"from":"writer","to":"reviewer"}
      ]
    },
    "input": {"topic": "Production multi-agent systems in 2026"}
  }' | python3 -m json.tool

Live demo — 7 multi-agent scenarios

pip install httpx
python demo_multiagent.py --scenario all

Core Features

Fault Tolerance

from afmx import Node, RetryPolicy, CircuitBreakerPolicy, TimeoutPolicy

Node(
    name="external_api",
    handler="api_call",
    retry_policy=RetryPolicy(
        retries=5,
        backoff_seconds=1.0,
        backoff_multiplier=2.0,   # 1s → 2s → 4s → 8s → 16s
        jitter=True,
    ),
    circuit_breaker=CircuitBreakerPolicy(
        enabled=True,
        failure_threshold=5,
        recovery_timeout_seconds=60.0,
    ),
    fallback_node_id="api_fallback",
)

Framework Adapters

from afmx.adapters.langchain import LangChainAdapter
from langchain.tools import DuckDuckGoSearchRun

adapter = LangChainAdapter()
node = adapter.to_afmx_node(DuckDuckGoSearchRun(), node_id="search")

Built-in adapters: LangChain · LangGraph · CrewAI · OpenAI — all lazy-loaded.

Registering Handlers

from afmx.plugins import default_registry

@default_registry.agent("my_analyst")
async def analyst(node_input: dict, context, node) -> dict:
    return {"analysis": "...", "confidence": 0.87}

@default_registry.tool("web_search")
async def search(node_input: dict, context, node) -> dict:
    return {"results": await run_search(node_input["input"])}

Cognitive Execution Matrix

AFMX v1.2 introduces the Cognitive Execution Matrix — a 2D coordinate system that maps every node to a cognitive layer (what type of thinking) and an agent role (which domain role performs it).

                  ROLES (open, domain-specific)
                  OPS   ANALYST  QUANT  CLINICIAN  PARALEGAL
LAYERS  PERCEIVE   ■      □       □       □          □
(fixed) RETRIEVE   ■      □       ■       □          □
        REASON     □      ■       ■       ■          □
        PLAN       ■      □       □       ■          ■
        ACT        ■      □       ■       □          □
        EVALUATE   □      ■       □       ■          □
        REPORT     ■      □       □       □          □

Row axis = CognitiveLayer (fixed, universal, drives automatic LLM cost routing). Column axis = AgentRole (open string — any industry vocabulary).

from afmx import Node, NodeType, CognitiveLayer
from afmx.domains.finance import FinanceRole

# Finance domain node
node = Node(
    name            = "risk-scorer",
    type            = NodeType.AGENT,
    handler         = "risk_model",
    cognitive_layer = CognitiveLayer.REASON,    # → premium LLM auto-selected
    agent_role      = FinanceRole.RISK_MANAGER, # == "RISK_MANAGER"
)

# Healthcare domain node
from afmx.domains.healthcare import HealthcareRole
node = Node(
    name            = "diagnosis",
    type            = NodeType.AGENT,
    handler         = "diagnostic_model",
    cognitive_layer = "REASON",
    agent_role      = HealthcareRole.CLINICIAN,
)

# Custom domain — any UPPER_SNAKE_CASE string is valid
node = Node(
    cognitive_layer = "PLAN",
    agent_role      = "DISPATCHER",   # logistics domain
    ...
)

Built-in domain packs: tech · finance · healthcare · legal · manufacturing. Custom domains: register in 8 lines with DomainPack + domain_registry.

LLM cost routing (automatic)

The CognitiveModelRouter auto-selects models by cognitive layer:

PERCEIVE / RETRIEVE / ACT / REPORT  →  cheap model  (Haiku, gpt-4o-mini)
REASON   / PLAN     / EVALUATE      →  premium model (Opus, o3, gpt-4o)

Typical result: 60–90% LLM cost reduction on multi-agent workflows.

REST API

Dashboard

cd afmx/dashboard
npm install && npm run build   # served at /afmx/ui
npm run dev                    # hot-reload at localhost:5173

Pages: Overview · Executions · Live Stream · Run Matrix · Saved Matrices · Plugins · Cognitive Matrix · Domain Packs · Audit Log · API Keys

Run Matrix includes cross-domain templates: cognitive (SRE) · finance · healthcare · legal.

Observability

@bus.subscribe(EventType.NODE_FAILED)
async def on_fail(event):
    await alert_team(event.execution_id, event.data["error"])

Prometheus metrics at GET /metrics. WebSocket streaming at WS /afmx/ws/stream/{id}.

Agentability Integration

AFMX integrates with Agentability — captures confidence scores, reasoning chains, token costs, and conflict detection per node execution.

AFMX_AGENTABILITY_ENABLED=true
AFMX_AGENTABILITY_DB_PATH=agentability.db
python demo_agentability.py

Docker

docker build -t afmx:latest .
docker run -p 8100:8100 --env-file .env afmx:latest

# Full stack: AFMX + Redis + Prometheus
docker-compose up -d

Testing

pytest                              # 290+ tests
pytest tests/unit/ -v               # unit only
pytest tests/integration/ -v        # integration only
pytest --cov=afmx --cov-report=html # coverage report

Documentation

AFMX vs alternatives (March 2026)

Mental model: AFMX = how agents act. LangGraph = how agents think. They are complementary — AFMX can execute LangGraph graphs as nodes.

Contributing

See CONTRIBUTING.md. All contributions welcome.

License

Apache 2.0 — see LICENSE.

Enterprise features (multi-tenancy, SSO/OIDC, cryptographic execution integrity, distributed workers, cost governance, AFMX Cloud) available under a separate commercial license. See ENTERPRISE.md or contact support@inteleion.com.