ORCHESTRATE AI AGENTS LIKE A MAINFRAME

Multi-agent orchestration for the command line. Chain AI agents together with parallel execution, dependency management, and seamless handover between stages. Your pipelines, your rules.

TERMINAL://DEMO LIVE
root@pipeline:~$ agent-pipeline run design-workflow.yaml
[14:23:01] INIT Pipeline: design-workflow
[14:23:02] START Stage: product-owner
[14:23:58] SUCCESS Stage: product-owner [56.2s]
[14:23:59] PARALLEL Spawning 3 agents...
[14:24:00] START Stage: brutalist-designer
[14:24:00] START Stage: cyberpunk-designer
[14:24:00] START Stage: indie-game-designer
[14:25:12] SUCCESS All parallel stages complete
[14:25:13] ACCESS GRANTED

CORE FEATURES

[||]

PARALLEL EXECUTION

Run multiple AI agents simultaneously. Maximize throughput by executing independent stages in parallel across your pipeline.

{=>}

DEPENDENCY GRAPH

Define stage dependencies with precision. The pipeline engine resolves execution order and manages the flow automatically.

[>>]

AUTO HANDOVER

Context flows between agents seamlessly. Each stage receives outputs from its dependencies via structured handover files.

/*.yaml

YAML CONFIGS

Define pipelines declaratively. YAML-based configuration keeps your workflows version-controlled and reproducible.

[!!]

FAILURE STRATEGIES

Configure how the pipeline handles failures. Stop on error, continue with warnings, or implement custom recovery logic.

[~~]

TIMEOUT CONTROL

Set per-stage timeouts to prevent runaway agents. Kill stages that exceed their allocated execution window.

HOW IT WORKS

01
DEFINE YAML
02
RUN PIPELINE
03
AGENTS EXECUTE
04
COLLECT OUTPUT
pipeline.yaml VALID
# Multi-agent design pipeline
name: design-exploration
version: "1.0"

stages:
  product-owner:
    agent: ./agents/product-owner.md
    timeout: 120

  brutalist-designer:
    agent: ./agents/brutalist.md
    depends_on: [product-owner]
    parallel: true

  cyberpunk-designer:
    agent: ./agents/cyberpunk.md
    depends_on: [product-owner]
    parallel: true

  final-review:
    agent: ./agents/reviewer.md
    depends_on: [brutalist-designer, cyberpunk-designer]

USE CASES

> DESIGN EXPLORATION

Run multiple design agents in parallel to explore different aesthetic directions. Compare outputs and iterate quickly.

> CODE GENERATION

Chain code generation, review, and testing agents. Automated quality gates ensure only passing code moves forward.

> CONTENT PIPELINES

Research, write, edit, and fact-check with specialized agents. Each stage refines the output of the previous.

> DATA PROCESSING

Extract, transform, and analyze data with agent chains. Parallel processing handles large datasets efficiently.

QUICK START

$ npm install -g agent-pipeline
TERMINAL://QUICKSTART
$ npm install -g agent-pipeline
$ agent-pipeline init my-pipeline
$ agent-pipeline run my-pipeline/pipeline.yaml
[OK] Pipeline initialized
[OK] Configuration valid
[OK] Executing stages...

JACK IN

Open source. MIT licensed. Ready for production.
Join the network of developers orchestrating the future.