AIOC

aioc is a TypeScript SDK for agent systems where models can propose actions, but deterministic runtime components decide what is actually allowed to happen.

It is designed for applications that must retain full governance control over persistence, tracing, approval, and oversight concerns, instead of inheriting those decisions from the SDK.

Status

The current stable release is 0.2.3. The stable line started with 0.1.0.

The current documentation covers:

• stable 0.1.x runtime behavior you can use today
• accepted RFC material implemented in the current stable release
• implemented 0.2.x additions, including the Agent Harness Descriptor API
• draft or experimental RFC material for future stabilization work that is still evolving

Stable Scope

aioc 0.1.0 stabilizes the governance-first runtime model as a dependable foundation for high-accountability applications: core APIs are compatibility-managed, audit artifacts and replay workflows are dependable, and the project has been validated through real implementation use.

aioc 0.1.1 keeps that stable model and adds thread-history helpers, the run-output stream adapter, and provider-specific instruction-role documentation.

aioc 0.1.2 adds approval evidence helpers for application-owned approval workflows.

aioc 0.2.x adds runtime utilities and the Agent Harness Descriptor API. The descriptor contract is part of the supported 0.2.x surface for controlled configuration and evaluation workflows; future 0.x changes should include migration guidance.

aioc 0.2.2 completes the descriptor instruction-composition surface with reusable instruction_parts, ordered instructions_sequence, and boolean where gates.

aioc 0.2.3 adds RFC-0010 policy composition helpers for exact-name tool and handoff policy dispatch without changing runtime enforcement semantics.

Install

npm install @axiastudio/aioc

What You Get Today

• deterministic policy gates for tools and handoffs
• RunRecord audit artifacts with prompt snapshots and request fingerprints
• replay and comparison utilities for non-regression analysis
• provider setup helpers for Mistral and OpenAI
• thread-history helpers for application-owned conversation state
• a run-output stream adapter for UI deltas plus final run data

Technical Focus

The project currently focuses on three technical areas:

• controlled execution through policy gates for tools and handoffs
• structured auditability through RunRecord, prompt snapshots, and request fingerprints
• verifiable iteration through replay and comparison utilities

What This Site Covers

• a concise technical introduction to the runtime model
• the key public concepts you need to build with aioc
• a compact manual reference for the most important public APIs
• run-record and governance-oriented documentation in a navigable format

Source of Truth

The normative governance documents remain in the repository root under /docs.

This site is a documentation app living in /apps/aioc-docs and imports those documents into a browsable Starlight site.

Where To Go Next

• Start with Quickstart for the minimal setup path.
• Read Governance-First Design for the design position behind the runtime.
• Use Reference for the current public runtime surface.
• Use Run Records if you are evaluating auditability, replay, and regression workflows.
• Use Governance for RFCs, contracts, and privacy baseline documents.