StackA2A
enterprisecrewaipython

Apflow

43

by aipartnerup

The core of `aipflow` is **task orchestration and execution specifications**. It provides a unified task orchestration framework that supports execution of multiple task types. The core is **pure orchestration** with no LLM dependencies - CrewAI support is optional.

7 starsUpdated 2026-02-18NOASSERTION
Quality Score43/100
★ Community
20
◷ Freshness
100
✓ Official
30
⚡ Skills
10
⬡ Protocol
40
🔒 Security
20

Getting Started

1Clone the repository
$ git clone https://github.com/aipartnerup/apflow
2Navigate to the project
$ cd apflow
3Install dependencies
$ pip install -r requirements.txt
4Run the agent
$ python main.py

Or connect to the hosted endpoint: https://aipartnerup.com/

README

apflow

Distributed Task Orchestration Framework

apflow is a distributed task orchestration framework that scales from a single process to multi-node clusters. It provides a unified execution interface with 12+ built-in executors, A2A protocol support, and automatic leader election with failover.

Start standalone in 30 seconds. Scale to distributed clusters without code changes.

Deployment Modes

Standalone (Development and Small Workloads)

pip install apflow

Single process, DuckDB storage, zero configuration. Ideal for development, testing, and small-scale automation.

from apflow.core.builders import TaskBuilder
from apflow import TaskManager, create_session

db = create_session()
task_manager = TaskManager(db)
result = await (
    TaskBuilder(task_manager, "rest_executor")
    .with_name("fetch_data")
    .with_input("url", "https://api.example.com/data")
    .with_input("method", "GET")
    .execute()
)

Distributed Cluster (Production)

pip install apflow[standard]

PostgreSQL-backed, leader/worker topology with automatic leader election, task leasing, and horizontal scaling. Same application code -- only the runtime environment changes.

# Leader node
APFLOW_CLUSTER_ENABLED=true \
APFLOW_DATABASE_URL=postgresql+asyncpg://user:pass@db:5432/apflow \
APFLOW_NODE_ROLE=auto \
apflow serve --port 8000

# Worker node (on additional machines)
APFLOW_CLUSTER_ENABLED=true \
APFLOW_DATABASE_URL=postgresql+asyncpg://user:pass@db:5432/apflow \
APFLOW_NODE_ROLE=worker \
apflow serve --port 8001

Add worker nodes at any time. The cluster auto-discovers them via the shared database.

Installation Options

Extra Command Includes
Core pip install apflow Task orchestration, DuckDB storage, core executors
Standard pip install apflow[standard] Core + A2A server + CLI + CrewAI + LLM + tools
A2A Server pip install apflow[a2a] A2A Protocol server (HTTP/SSE/WebSocket)
CLI pip install apflow[cli] Command-line interface
PostgreSQL pip install apflow[postgres] PostgreSQL storage (required for distributed)
CrewAI pip install apflow[crewai] LLM-based agent crews
LLM pip install apflow[llm] Direct LLM via LiteLLM (100+ providers)
SSH pip install apflow[ssh] Remote command execution
Docker pip install apflow[docker] Containerized execution
gRPC pip install apflow[grpc] gRPC service calls
Email pip install apflow[email] Email sending (SMTP)
All pip install apflow[all] Everything

Built-in Executors

Executor Purpose Extra
RestExecutor HTTP/REST API calls with auth and retry core
CommandExecutor Local shell command execution core
SystemInfoExecutor System information collection core
ScrapeExecutor Web page scraping core
WebSocketExecutor Bidirectional WebSocket communication core
McpExecutor Model Context Protocol tools and data sources core
ApFlowApiExecutor Inter-instance API calls for distributed execution core
AggregateResultsExecutor Aggregate results from multiple tasks core
SshExecutor Remote command execution via SSH [ssh]
DockerExecutor Containerized command execution [docker]
GrpcExecutor gRPC service calls [grpc]
SendEmailExecutor Send emails via SMTP or Resend API [email]
CrewaiExecutor LLM agent crews via CrewAI [crewai]
BatchCrewaiExecutor Atomic batch of multiple crews [crewai]
LLMExecutor Direct LLM interaction via LiteLLM [llm]
GenerateExecutor Natural language to task tree via LLM [llm]

Architecture

                    +--------------------------+
                    |    Client / CLI / API     |
                    +------------+-------------+
                                 |
              +------------------+------------------+
              |                  |                   |
    +---------v--------+ +------v------+ +----------v--------+
    |   Leader Node     | | Worker Node | |   Worker Node      |
    |  (auto-elected)   | |             | |                    |
    |  - Task placement | |  - Execute  | |  - Execute         |
    |  - Lease mgmt     | |  - Heartbeat| |  - Heartbeat       |
    |  - Execute        | |             | |                    |
    +---------+--------+ +------+------+ +----------+--------+
              |                  |                   |
              +------------------+------------------+
                                 |
                    +------------v-------------+
                    |  PostgreSQL (shared)      |
                    |  - Task state             |
                    |  - Leader lease           |
                    |  - Node registry          |
                    +--------------------------+

Standalone mode uses the same architecture with a single node and DuckDB storage.

Documentation

Full documentation: flow-docs.aipartnerup.com

Contributing

Contributions are welcome. See Contributing Guide for setup and guidelines.

License

Apache-2.0

Capabilities

StreamingPush NotificationsMulti-TurnAuth: none
agent-frameworkai-agentsautomation-task-schedulercrewaidistributed-systemslitellmmcpscheudlerstandalonetask
View on GitHub