Ploston¶

LLM plans. Ploston executes.

The Problem¶

Today's AI agents orchestrate their own tool calls. The LLM decides which tool to call, interprets the result, decides the next step, and repeats—burning tokens and hallucinating along the way.

This creates five systemic failures:

No enterprise will deploy this in production. No developer wants to debug it.

The Solution¶

Ploston moves orchestration out of the LLM and into a deterministic runtime.

flowchart LR
    subgraph TODAY["TODAY"]
        A["LLM = planner + executor<br/>(fragile, expensive)"]
    end
    subgraph WITHPLOSTON["WITH Ploston"]
        B["LLM = planner"]
        C["Ploston = executor<br/>(deterministic, fast)"]
    end

You define workflows in YAML. Ploston exposes them as MCP tools. When your agent needs to scrape a website, transform the data, and publish it—it makes one call to your workflow. Ploston handles the rest.

Same inputs. Same outputs. Every time.

How It Works¶

flowchart TB
    Agent["Agent (Claude, GPT, etc.)"]
    Agent -->|"MCP call: workflow:scrape-and-publish"| Ploston

    subgraph Ploston["Ploston"]
        direction TB
        S1["Step 1: fetch_url"]
        S2["Step 2: extract_data"]
        S3["Step 3: validate_schema"]
        S4["Step 4: publish_to_kafka"]
        S1 --> S2 --> S3 --> S4

        Features["✓ Deterministic execution<br/>✓ Full trace and audit log<br/>✓ Built-in retry and errors"]
    end

    Ploston --> Result["Result returned to agent"]

The ✓ marks indicate guarantees Ploston provides: deterministic execution (same inputs → same outputs), complete audit trails, and automatic retry/error handling.

The agent doesn't orchestrate. It delegates to infrastructure that executes reliably.

Learn how this works →

Quick Start¶

# Install the CLI
pip install ploston-cli

# Run a workflow
ploston run workflows/hello-world.yaml

# Start as MCP server (connect to Claude Desktop)
ploston serve

Get Started →

What You Get¶

OSS (Available Now)¶

• Workflow Engine — YAML-defined, deterministic execution
• MCP Native — Expose workflows as tools agents can call
• Python Code Steps — Sandboxed execution with 7-layer security
• CLI — Manage, test, and run workflows locally
• Telemetry — Execution traces and structured logging

Enterprise (Coming Soon)¶

• Governance — RBAC, ABAC, policy enforcement
• Advanced Workflows — Parallel execution, human approval, compensation
• Pattern Mining — Detect repeated tool chains, suggest workflows
• Workflow Synthesis — LLM-generated workflows from observed patterns
• Cost Accounting — Track token savings, report ROI

See Roadmap →

Why Ploston?¶

Ploston is not an API gateway. It's not a workflow engine. It's not an agent framework.

It's the execution layer that makes agent systems production-ready.

Learn why this matters →

Architecture¶

flowchart TB
    Agent["AI Agent / Client"]

    Agent -->|"MCP Protocol / REST API"| Ploston

    subgraph Ploston["Ploston"]
        direction TB
        subgraph Components[" "]
            direction LR
            WE["Workflow<br/>Engine"]
            TR["Tool<br/>Registry"]
            MF["MCP Frontend<br/>(stdio or HTTP)"]
        end
        Sandbox["Python Sandbox (Code Steps)"]
        Components --> Sandbox
    end

    Ploston -->|"MCP Protocol"| External

    subgraph External["External MCP Servers"]
        Tools["filesystem, fetch, databases, custom tools, etc."]
    end

Key Concepts¶

Documentation¶

Getting Started¶

Concepts¶

Guides¶

Reference¶

Examples¶

System Requirements¶

• Python: 3.12 or higher
• OS: macOS, Linux, Windows
• Memory: 512MB minimum, 1GB recommended

Getting Help¶

• GitHub Issues: Report bugs or request features
• Discussions: Ask questions and share ideas

License¶

Ploston is released under the Apache 2.0 License.