OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
Multi-Agent Systems

Flock — whiteducksoftware

Flock is a Python framework for coordinating AI agents through a shared blackboard. Developers define typed artifacts that agents consume and publish; those subscriptions trigger workflow steps and allow independent agents to run in parallel. The project also describes a real-time dashboard, OpenTelemetry tracing and decision-model routing.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Flock — whiteducksoftware — OpenViST project profile in Multi-Agent Systems.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use Flock when a task benefits from multiple agents exchanging structured artifacts, especially when parts can run in parallel and later steps depend on specific outputs.

Who it is for

  • Python developers building workflows that divide work among multiple AI agents.
  • Teams that want agent handoffs expressed as typed data contracts.
  • Developers who need to inspect workflow activity or route tasks using structured decisions.

Documented uses

  • Define typed inputs and outputs for agents using Pydantic models.
  • Run independent agent analyses in parallel and wait for required results before a downstream step.
  • Route tasks based on typed decisions and thresholds.
  • Inspect activity in a live dashboard and trace runs with OpenTelemetry.
Before you choose

Requirements & limits

Flock is an MIT-licensed Python framework distributed as the `flock-core` package. The quick-start example requires Python 3.12 or later and specifies an OpenAI API key and model setting.

  • Workflow logic is expressed through artifact subscriptions rather than a manually wired graph.
  • Pydantic models provide the documented typed contracts and output validation.
  • The README quick start uses OpenAI; decision-model support is described separately and names Microsoft-Decision-1, OpenAI Decisions, TypeSafe Jev and a local Clef model.
  • A dashboard and OpenTelemetry tracing are identified as built-in capabilities.
  • The repository is public, MIT-licensed and not marked archived.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Build and run a typed, parallel agent review workflow

Publish a code submission, run two analysis agents in parallel, and have a reviewer consume both results.

Before you start

  • Python 3.12 or later
  • Install the flock-core package with pip install flock-core (or uv add flock-core).
  • Set OPENAI_API_KEY and DEFAULT_MODEL to the values shown in the Quick start; the workflow initializes Flock with openai/gpt-4.1.
  • Use Pydantic models for the typed artifacts.

Steps

  1. Define typed Pydantic artifacts for the code submission, bug analysis, security analysis, and final review; the README example decorates each model with @flock_type.

  2. Create Flock("openai/gpt-4.1") and declare bug_detector and security_auditor as consuming CodeSubmission and publishing their respective analysis artifacts.

  3. Declare reviewer as consuming both BugAnalysis and SecurityAnalysis and publishing FinalReview; this subscription acts as an AND gate.

  4. Publish a CodeSubmission, call await flock.run_until_idle(), then retrieve FinalReview artifacts from the store and print the verdict and action items.

Expected outcome

The two analysis agents run in parallel, and the reviewer runs after both analyses are available; typed outputs are validated by Pydantic.

Guide limitations

  • The example requires the configured OpenAI model and API key; no provider setup beyond the environment variables is described.
  • The documented workflow runs until idle and retrieves results from the store; it does not document deployment.
Documentation 1 ↗
Source-based practical guide

Serve the review workflow with the dashboard

Use the documented dashboard-serving option to view agent activity and publish artifacts from a browser.

Before you start

  • A Flock workflow configured as in the Quick start, including the package and model environment settings.

Steps

  1. Define the typed artifacts and agent subscriptions for the review workflow as shown in the Quick start.

  2. In the asynchronous main() function, replace the publish-and-run body with await flock.serve(dashboard=True).

  3. Run the application using the Python entry point that calls asyncio.run(main()).

Expected outcome

Flock serves the workflow with the dashboard, where the README says agents can be watched and artifacts published from the browser.

Guide limitations

  • The README does not provide additional dashboard configuration, access controls, or deployment instructions.
Documentation 1 ↗
COMPARE OPTIONS

Compare options for Agent orchestration

These projects document a common task and share a project role. This is a source-based comparison, not an execution benchmark or a guarantee of interchangeability.

ProjectDocumented purposePlatformAccess and requirements
Flock — whiteducksoftwareDefine typed inputs and outputs for agents using Pydantic models.CLI, Self-hosted, ServerFlock is an MIT-licensed Python framework distributed as the `flock-core` package. The quick-start example requires Python 3.12 or later and specifies an OpenAI API key and model setting.
MassgenRun multiple AI agents against the same user task.CLIThe repository documents installation through pip or uv and requires Python 3.11 or later. Setup can configure API keys for OpenAI, Anthropic, Google and xAI; credentials are needed for the selected model backends.
Langchain Multi Agents BoilerplateRoute incoming football questions to model and agent workflows according to task complexity.WebThe repository is public and specifies an MIT license. Running the application requires OpenAI and Groq API keys and a Redis server.
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
12 October 2026
Original website title
whiteducksoftware/flock · Official homepage title · retained in its source language
How we review sourcesSuggest a correction