Holon

Open-source AI assistants on your own computer or server. Assign tasks, track progress, and let them follow up on ongoing work.

THE HOLON BLOG

Work that keeps going.

The ideas behind Holon, the design that supports them, and the people putting agents to work.

START HERE · PRODUCT

What is Holon? Let agents keep working in your environment.

Give recurring work to agents with ongoing roles. A tour of the local workbench, from continuous review to remote development and team use.

Meet Holon

Design & practice

How it works. How to use it.

HANDS-ON GUIDE

From prompt to finished work

A repeatable workflow for giving an agent a real repository task, letting it work past the prompt, and checking the result when it comes back.

Follow the workflow
DESIGN

WorkItems: separating work state from conversations

An API migration and two PRs show why goals, progress, and waiting conditions need a home of their own.

Explore the design
HANDS-ON GUIDE

Build a reviewer that follows PRs through

Set up the code-reviewer template, agree on permissions, and subscribe to PRs. Follow revisions and CI through to the merge decision.

Set up continuous review
TEAM STORY

From personal AI tools to shared team agents

A small team's 131 days of recorded usage, wake-source breakdown, device investigations, and hands-on testing show where shared agents fit—and where people still step in.

Read the team story
ARCHITECTURE

From Request-Response to Event-Driven: Designing the Kernel for Long-Lived Agents

Why while(true) loops fail in production engineering. Drawing from OS history, we explore durable wait-wake scheduling and causal context design.

Read the architecture
ENGINEERING PRACTICE

How to Guarantee Quality in AI Delivery Pipelines: Four Layers of Verification and Acceptance in Real-World Engineering

Code generation is faster than ever, but software nobody dares deploy remains the real bottleneck. Drawing from an engineering team adopting Holon across cross-platform apps and hardware protocols, this article explores four pragmatic verification layers.

Read the guide
SYSTEM DESIGN

Why your agent gets dumber with more skills (and how to fix it)

Adding 30 skills to a single agent triggers context dilution, instruction conflicts, and safety risks. Here is how Holon isolates skills across long-lived roles.

Read the analysis

Try it with one responsibility.

Install Holon, connect a provider, and give your first agent a bounded job.

Get started
What is Holon? Multiple agents working in your environment2026-09-13 · Assign recurring work to established roles, let agents follow up in the background, and reconnect when needed. An ongoing review example introduces Holon's local workbench for remote development and team collaboration.Holon's WorkItem architecture: separating work state from conversation2026-09-13 · Follow an interface migration from launch to retirement to see how WorkItem differs from Goal and Loop: it keeps objectives, progress, external dependencies, and delivery together across release cycles.Beyond a one-time review: build a reviewer that follows PRs with Holon2026-09-13 · Create a reviewer with built-in working rules and Skills, set its responsibilities and merge permissions, and have it discover PRs and follow fixes and CI updates.Why your agent gets dumber with more skills (and how to fix it)2026-11-08 · Adding 30 skills to a single agent triggers context dilution, instruction conflicts, and safety risks. Here is how Holon isolates skills across long-lived roles.From personal AI tools to team collaboration: an agent-native case study2026-09-13 · How a small team moved from individual AI tools to shared agents, with 131 days of retained usage records, wake-source breakdown, device investigations, and a hands-on test of silent piano keys.From prompt to finished work: a practical Holon workflow2026-10-06 · A repeatable workflow for giving an agent a real repository task, letting it work past the prompt, and checking the result when it comes back.From Request-Response to Event-Driven: Designing the Kernel for Long-Lived Agents2026-10-09 · Why while(true) loops fail in production software engineering. Drawing from OS history—from busy-waiting to hardware interrupts and epoll—we explore durable wait-wake scheduling, decoupled work state, and causal contexts for long-lived autonomous agents.How to Guarantee Quality in AI Delivery Pipelines: Four Layers of Verification and Acceptance in Real-World Engineering2026-10-14 · Code generation is faster than ever, but software nobody dares deploy remains the real bottleneck. Drawing from an engineering team adopting Holon across cross-platform apps and hardware protocols, this article explores four pragmatic verification layers: role separation, automated incident triage, continuous instance patrols, and self-healing E2E testing.