Skip to content

ESF

Self-hosted software factory running coding agents in microVMs with verification gates; offers optional DSPy/Jev task intake advice.

ServicebetaMIT

Overview

ESF (Engineering Software Factory) is a self-hosted system that runs coding agents in isolated CubeSandbox microVMs, verifies their changes and preserves the patch and execution evidence, with Temporal coordinating the workflow and sandbox cleanup. It is a fork of Machinist, keeping its CLI, local control plane and web UI and adding factory orchestration, operator-configured harnesses including an opt-in Pi harness, deterministic verification gates, durable artefacts, TLS and API-key security, egress boundaries and a behaviour monitor.

Its link to decision models is small: the README lists an optional DSPy/Jev intake component that provides advisory typed task advice and secure TypeSafe credential setup. ESF itself states that agent success alone is not factory success and that it produces changes for review rather than deciding what ships. Version 0.6.1 is a release candidate.

What the decision model decides

TypeQuestionAnswers
ChoiceAdvisory typed classification of an incoming task at intake(inferred)—

How it works

  1. Task enters via Temporal
  2. Optional DSPy/Jev intake gives advisory task advice
  3. Agent runs in a CubeSandbox microVM
  4. Deterministic gates verify; patch and evidence are kept

Strengths

  • Strong isolation and evidence retention
  • Jev intake is optional and advisory only

Caveats

  • Jev plays a minor optional role
  • Release candidate; requires existing CubeSandbox and Temporal deployments

Getting started

Review docs/production-readiness.md and the release procedure before deploying against an existing CubeSandbox and Temporal setup.

At a glance

Type
Service
Status
beta
Language
Python
Runtime
self-hosted service (Temporal + CubeSandbox)
Has code
yes
Has tests
yes
License
MIT

Use cases and integrations

How curated lists describe it

  • Self-hosted software factory: coding agents in microVMs, changes verified, and the patch and execution evidence kept, with Temporal coordinating the workflow.

    MrJev/awesome-jev

Frequently asked questions

Where is Jev used?

In an optional DSPy/Jev intake step that gives advisory typed task advice.

Does ESF decide what ships?

No, it produces verified changes for human review.

Where do agents run?

In isolated CubeSandbox microVMs.

Related projects

  • Lean Pi supervisor: zero-token regex guardrails, done-check verification, per-turn routing and a Jev/Clef semantic slop gate on diffs.

    ProjectbetaTypeScript
  • Pi guardrails on pi-typesafe that steer the agent: Jev checks rule breaks, irreversible actions, stuck loops and unverified done claims.

    HookbetaTypeScript
  • Auditable Zig terminal agent pairing Codex for coding with Jev typed preflight and postflight checks under deterministic policy gates.

    CLIproductionZig
  • Claude Code plugin with nine Jev-judged gates for rules, scope, intent, completion and claims that escalate but never approve.

    HookbetaTypeScript
  • Claude Code plugin that scores a turn's redacted diff on eight Jev risk axes at Stop and asks Claude to reinvestigate once if high.

    ProjectbetaGo
  • Risk gate for AI coding agents: Jev decides at five layers and Claude Code hooks enforce deny or ask deterministically.

    ProjectbetaJavaScript