RuleOak Core v1.0

Build AI agents that can act, but cannot act recklessly.

RuleOak gives developers an early runtime foundation for governed AI workflows: policy-bound actions, evidence-backed recommendations, approval gates, audit-style records, and deny-by-default sandbox boundaries.

Explore Core v1.0 View GitHub

npm install
npm run launch
npm run demo
npm run report:view

v1.0 first release: early runtime, sandbox foundation, two demos, launch UX, HTML reports, local viewer, tests, and CI.

Why RuleOak

Agent autonomy needs accountable boundaries.

RuleOak is for builders who want useful AI workflows without giving the model unchecked authority over tools, files, systems, or decisions.

Policy

Define allowed, blocked, and approval-gated actions outside the prompt.

Evidence

Ground recommendations in logs, documents, metrics, notes, or cited sources.

Approval

Pause before risky or external actions so humans stay accountable.

Audit

Record runs, evidence, policy decisions, approvals, and outputs for review.

Launch path

From clone to governed workflow in one command.

The v1.0 package includes a guided first-run experience, workflow chooser, templates, one-page HTML reports, and a local report viewer.

git clone git@github.com:ruleoak/ruleoak-core.git
cd ruleoak-core
npm install
npm run launch

Two demos

One pattern, multiple workflow shapes.

Technical Consultant Demo

Case inputs, evidence, probable cause, recommended action, approval boundary, and audit-style report.

Research Brief Demo

Sourced claims, confidence labels, known unknowns, recommendation, and publication approval boundary.

Sandbox foundation

Deny by default. Approve what matters.

RuleOak Core v1.0 includes filesystem, network, command, and tool policy guards with automated tests. It is a sandbox foundation, not an externally security-reviewed sandbox yet.

Filesystem

Workspace boundary, denylisted secrets, and read/write allowlists.

Network

External network denied by default; localhost can be allowed for local runners.

Commands

Commands are classified as allow, deny, or approval-required.

Tools

Registered tools are policy-checked before use.