Skip to content
AIOC Docs

Examples

All example commands are run from the repository root.

Most live examples use AIOC_EXAMPLE_PROVIDER=openai or AIOC_EXAMPLE_PROVIDER=mistral plus the matching provider API key. Optionally set AIOC_EXAMPLE_MODEL to override the default live model.

Start Here

Section titled “Start Here”
  • npm run example:hello
  • npm run example:policy
  • npm run example:approval-required
  • npm run example:tool-policy
  • npm run example:policy-composition

These are the compact examples intended for first reading. They cover the smallest useful agent run, policy gates, approval-required results, tool execution, and exact-name policy composition.

Approval and Audit

Section titled “Approval and Audit”
  • npm run example:approval-evidence
  • npm run example:run-record

example:approval-evidence shows how an external approval can be projected into policy context and reevaluated deterministically. example:run-record shows how to persist a redacted RunRecord through a sink.

RunRecord Utility Snippets

Section titled “RunRecord Utility Snippets”
  • npm run example:rru:01-extract
  • npm run example:rru:02-compare
  • npm run example:rru:03-replay-strict
  • npm run example:rru:04-replay-hybrid

These examples are deterministic and runnable without a live provider. They are small reference snippets for extracting tool calls, comparing records, and replaying recorded runs.

Harness Descriptor Examples

Section titled “Harness Descriptor Examples”
  • npm run example:harness-rerun
  • npm run example:harness

example:harness-rerun is the compact descriptor example: it configures OpenAI from OPENAI_API_KEY, declares the model in inline YAML, records a source run, and replays it against a modified harness while mocking a newly introduced tool output.

example:harness is the fuller descriptor example. It demonstrates reusable instruction_parts, ordered instructions_sequence, boolean where gates, tool registry binding, and descriptor hashing.

Advanced Workflows

Section titled “Advanced Workflows”
  • npm run example:non-regression
  • npm run example:run-regression
  • npm run example:run-regression-judge

example:non-regression compares RunRecord outputs across two harness versions. It uses a live provider and is educational rather than deterministic.

example:run-regression records a baseline RunRecord, runs a modified OpenAI-backed harness through runRegressionSuite(...), and prints the deterministic comparison summary without using a judge.

example:run-regression-judge extends the same age-adapted regression flow with createRunRegressionJudge(...). The judge package builds the bounded judge input and prompt; the example supplies the actual OpenAI-backed model call.

The regression CLI artifact example lives in packages/aioc-regression-cli/examples/age-adapted/. It shows the same kind of workflow using stored RunRecord JSON, descriptor YAML, an expectation file, and an application-owned .mjs adapter.

Optional LangChain Interoperability

Section titled “Optional LangChain Interoperability”

Optional LangChain examples live in /examples/langchain and use an isolated package.json, so LangChain dependencies do not become runtime dependencies of @axiastudio/aioc.

They demonstrate three composition patterns:

  • aioc-first, LangChain-extended: aioc owns the governed agent run while LangChain provides OSS components behind aioc tools. The RAG example mirrors the LangChain RAG tutorial domain and uses a LangChain retriever inside a policy-gated aioc tool.
  • LangGraph-orchestrated, aioc-governed: LangGraph owns workflow orchestration while selected graph nodes call aioc.run(...) for sensitive capability execution. The LangGraph example mirrors the calculator quickstart domain and routes the governed step through aioc policy and RunRecord capture.
  • LangGraph-orchestrated, aioc-recorded: LangGraph owns workflow orchestration while a local RFC-0013 prototype wraps the compiled graph and emits a graph-level RunRecord.

For governed execution, execution-impacting capabilities should cross the aioc governance boundary. LangChain supplies breadth and orchestration; aioc supplies default-deny authorization, deterministic policy decisions, and portable audit evidence.

Canonical Guide

Section titled “Canonical Guide”

For the curated learning path, see the repository document:

  • /docs/CANONICAL-EXAMPLES.md

For the approval model and policy-reevaluation flow, see:

  • /approval-flows/