Skip to content

AI Workflow Standard

LRIL AI
AI Workflow Standard

Define. Govern.
Execute. Observe.

A standard, three native SDKs, and Praxis—the AIWS workflow engine—for building governed, observable AI workflows across agents, tools, humans, and systems.

  • SDK0.3.0
  • ProposalAIWS-001 · 0.4
  • LanguagesTS · PY · RS
A luminous AI workflow system with six equally connected modules surrounding a central core.
AIWSshared semantics
Standardcontracts · schema
SDKsTypeScript · Python · Rust
Praxisworkflow engine · state · recovery
Evidenceevents · validation
Agents & toolsmodels · adapters
Human controlapprove · intervene

The complete stack

One system. Three layers.

Shared semantics connect what you define, what developers build, and what the runtime executes.

Governed execution

From intent to evidence.

Every stage has a defined shape, authority boundary, and durable record.

  1. 01Trigger
  2. 02Work order
  3. 03Agent
  4. 04Handoff
  5. 05Verification
  6. 06Evidence
Controlled · inspectable · recoverable · auditable

Human authority

Agents work.
Humans remain in control.

Approval gates, permissions, intervention, recovery, and uncertainty handling are part of the workflow—not exceptions bolted on afterward.

Explore authorization and control
WORK ORDER · WO-1842 Awaiting approval
A
AGENT REQUESTApply verified implementation
02:14
Authority
Propose only
Evidence
7 artifacts attached
Risk
Repository write
PauseResumeOverrideReplace agent

Native SDKs

Use the language you already work in.

Three independent libraries implement the same finite-v1 profile and share exact conformance scenarios.

# Install the native TypeScript SDK
npm install ./aiws-sdk-0.3.0.tgz

# Run the verified first workflow
node examples/guide/typescript/first-run.ts
Explore SDK documentation

Why a standard?

AI workflows need a common language.

Agents, orchestrators, tools, and runtimes describe work differently. That makes governance, portability, recovery, and trustworthy evidence harder than they need to be.

AIWS defines shared semantics for how AI-powered work is described, executed, controlled, resumed, verified, and recorded.

01

Interoperable

Workflows do not belong to one model, agent, vendor, or runtime.

02

Durable

Execution survives pauses, retries, failures, and long-running work.

03

Human-centered

People can inspect, approve, intervene, and retain authority.

04

Observable

Every run produces structured state, events, evidence, and history.

05

Governed

Permissions and policies constrain what agents are allowed to do.

06

Verifiable

Acceptance conditions and evidence—not a completion claim—close the work.

Start building

Build workflows agents
can be trusted to run.

Start with the standard, choose an SDK, and run your first governed workflow.