data-machine

What Data Machine does

Data Machine is an agentic workflow automation layer for WordPress. It turns a WordPress site into an agent runtime with persistent identity, memory, pipelines, abilities, and tools that AI agents use to operate autonomously. The plugin builds on the Agents API project for generic agent runtime contracts and durable agent primitives, while Data Machine owns the WordPress automation product layer: pipelines, flows, jobs, handlers, tools, abilities, memory files, system tasks, request assembly, and the admin and CLI surfaces. You can try it in a browser-only WordPress Playground that boots WordPress, installs Data Machine plus its substrate, and drops you into the admin screen. Pipelines are multi-step workflows that fetch content, process it with AI, and publish it anywhere. The Abilities API exposes typed, permissioned functions such as datamachine/upload-media and datamachine/validate-media that agents and extensions call. Agent memory uses layered markdown files, with SOUL.md and MEMORY.md in the agent layer and USER.md in the user layer, injected into every AI context. Multiple agents can run with scoped pipelines, flows, jobs, and filesystem directories, and agents can schedule their own recurring tasks using flows, prompt queues, and Agent Pings, which lets a site keep working between human visits. The Playground badge means a reviewer can see the admin without setting up a server, which shortens the path from curiosity to a working test of the automation layer.

Pipelines and execution model

A pipeline defines a workflow template with steps such as fetch, AI, and publish. A flow configures and schedules a pipeline, and a job tracks each execution with its artifacts, retries, status, and undo support. The engine processes work as a bounded cycle: it claims due work, advances jobs, and exits, rather than running one unbounded request. Long-running automation uses small repeatable units named cycle, drain, and worker. A cycle is one bounded scheduler pass that claims due work and advances jobs. A drain is a CLI command that repeatedly runs due work until the queue is empty or a budget is reached. A worker is a runtime wrapper for ongoing background processing that runs cycles under configured limits. A retention task cleans old logs, jobs, processed items, files, and chat sessions. The README lists entry points such as `wp datamachine cycle`, `wp datamachine drain`, `wp datamachine worker`, and `wp datamachine retention`. One agent can run in three modes with the same identity and memory but different guidance and tools: pipeline mode for automated workflows, chat mode in wp-admin with management tools, and system mode for background infrastructure tasks like alt text and meta descriptions, so the agent wears different hats for different kinds of work. The cycle, drain, and worker model avoids one unbounded request, which keeps long jobs from blocking the web request that started them and improves stability under load.

Abilities, tools, and governance

Tools are resolved at runtime by policy classes instead of hardcoded allow-lists. Data Machine combines tool sources, mode policy, handler adjacency, memory policy, and action policy to decide what an agent can see and what needs approval. When an operation needs human approval it becomes a pending action backed by the Agents API approval vocabulary and Data Machine storage. Pending actions can be inspected and resolved through REST, abilities, chat tools, and CLI. The Abilities API registers typed, permissioned functions through WordPress's Abilities API, and extensions and agents consume them instead of reaching into internals. Core abilities include query-posts, publish-wordpress, update-wordpress, generate-alt-text, generate-meta-description, and run-flow, with more across pipelines, flows, jobs, memory, media, SEO, email, and infrastructure. Social publishing, workspace, GitHub, and business integration abilities live in extension plugins such as data-machine-socials, data-machine-code, and data-machine-business. Agent bundles package a recipe as manifest metadata, memory, pipelines, flows, prompts, rubrics, tool policies, and auth references. Content abilities accept a content format of markdown, html, or blocks, and Data Machine stores content in the post type's canonical format, which keeps WordPress as the system of record. Approval as a pending action means a human can sit between the agent and a risky write, which fits a site that publishes public content and needs oversight on changes.

Surface area and requirements

Data Machine exposes a wide WP-CLI surface plus a REST API under datamachine/v1. The CLI covers agents, ai, analytics, alt-text, auth, batch, blocks, chat, cycle, drain, email, external, flows, handlers, image, jobs, links, logs, memory, meta-description, pending-actions, pipelines, posts, processed-items, retention, settings, step-types, system, taxonomy, worker, and more, with `wp help datamachine` as the authoritative list. The REST API includes execute, trigger, pipelines, flows, jobs, logs, chat, files, agents, agent-ping, handlers, step-types, providers, tools, settings, system, auth, email, users, analytics, links, and pending actions. Extensions add socials, code, business, editor, frontend chat, chat bridge, events, and recipes. AI providers are configured through wp-ai-client with a global default and per-mode overrides for pipeline, chat, and system. Requirements are WordPress 6.9 or later for the Abilities API, PHP 8.2 or later, and a bundled Action Scheduler. Data Machine is licensed under GPL-2.0 and ships with a 1.0 baseline that starts from a single canonical schema and runtime contract, which gives downstream builders a stable foundation to extend. The combination of a wide CLI, a REST API, and extension plugins means a site owner can start with chat and grow into scheduled pipelines without leaving the WordPress admin they already use. The 1.0 baseline gives extension authors a fixed contract to build against, which reduces the risk that a core change breaks a plugin that depends on the runtime surface.

Editorial conclusion

Data Machine is written in PHP, distributed under the GPL-2.0 license, and its repository was last updated on 2026-08-24.