holon
Local agent workbench whose decision crate borrows the System One state/questions request shape, starting with Cloudflare's Clef.
Overview
Holon is a local, Apache-2.0 licensed workbench for agents doing ongoing work rather than one-off prompts. It keeps goals, plans, progress and wait conditions in explicit WorkItems, so an agent can wait for a command result, a review or an external event and then resume in the right workspace. It offers a terminal UI and a web GUI, a daemon that keeps agents running after you disconnect, and explicit trust boundaries between operator input, external events and tool results.
Its relevance to this directory is architectural. According to the list summary, Holon keeps decisions in their own crate, and the adapter for its first decision backend, Cloudflare's Clef, notes that Clef uses the System One-shaped state/questions request. In other words, Holon reuses Jev's interface shape for typed decisions instead of inventing its own. The README itself focuses on the workbench and does not describe the decision layer in detail.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Other | Typed state/questions decisions routed through a pluggable decision backend (first backend: Clef)(inferred) | — |
How it works
- Agent works on a WorkItem in a local workspace
- When a typed judgment is needed, the decision crate builds a state/questions request
- The backend adapter (Clef) returns typed answers
- Agent continues, waits or wakes based on the result
Strengths
- Decision logic is isolated in its own crate behind a common request shape
- Reusing the System One schema makes backends swappable
- Local-first execution keeps work in real repositories and toolchains
Caveats
- The decision layer is a small part of a larger agent workbench and is not documented in the README
- First backend is Clef, not TypeSafe Jev
Getting started
`brew tap holon-run/tap && brew install holon`, then `holon onboard`, `holon daemon start` and `holon tui`.At a glance
- Type
- Library
- Status
- beta
- Language
- Rust
- Runtime
- daemon with TUI and web GUI
- Has code
- yes
- License
- Apache-2.0
Use cases and integrations
How curated lists describe it
Agent workbench: keeps decisions in their own crate and notes in the adapter for its first backend, Cloudflare's Clef, that "Clef uses the System One-shaped state/questions request" — the interface being borrowed rather than reinvented.
yibie/awesome-jev
Frequently asked questions
Is Holon a Jev project?
Not directly; it is an agent workbench whose decision crate uses the same state/questions request shape as Jev, with Clef as its first backend.
What is a WorkItem?
Holon's unit of ongoing work, holding goals, plans, progress and the conditions an agent is waiting on.
How do I install it?
Via Homebrew (holon-run/tap) or binaries from GitHub Releases.
Related projects
Nerve: asynchronous System-1 supervisory plugin for Hermes Agent judging relevance, completion and recovery with Jev, Laya or OpenJev.
ProjectbetaPythonOpen-weight 0.6B decision model on a Qwen3 backbone: state and typed questions in, a probability for every option out, nothing decoded.
ModelresearchPythonDeepSeek Harness plugin that registers a native jev_decide tool and declares its own policy for when the agent should ask Jev.
PluginbetaJavaScriptHermes skill that calls Jev as a typed tiebreaker, returning Choice, Score or Noul probabilities so the agent can act or escalate.
Agent skillbetaPythonDocker service exposing local Laya, Von, a CLM head or hosted Jev through one TypeSafe-compatible HTTP and MCP endpoint.
MCP serverbetaPythonnpm library that rate-limits, spend-caps, caches and redacts every Jev request made in a loop.
LibrarybetaJavaScript