Vesanor

Vesanor is an LLM function-calling reliability platform. It validates, monitors, and gates tool-call behavior across providers using a runtime SDK, deterministic contracts, replay evidence, and Governance Studio.

The main public path is:

1. wrap your OpenAI or Anthropic client with replay()
2. run real traffic through the wrapped client
3. review inferred behavior in Governance Studio
4. approve the runtime artifact when ready
5. let Vesanor monitor, advise, or enforce based on environment

No local YAML contracts are required for the default workflow.

Install

Install Vesanor with the provider SDK you use:

npm install vesanor openai

or:

npm install vesanor @anthropic-ai/sdk

Requires Node.js 18 or newer.

Quick Start

import OpenAI from "openai";
import { replay } from "vesanor";

const client = new OpenAI();

const session = replay(client, {
  apiKey: process.env.VESANOR_API_KEY,
  agent: "orders-bot",
});

const response = await session.client.chat.completions.create({
  model: "gpt-4o-mini",
  messages: [{ role: "user", content: "Process this order" }],
  tools: myToolDefinitions,
});

That is enough to start zero-config governance. Vesanor captures runtime evidence, builds review state for Governance Studio, and attaches approved runtime behavior after you approve.

What It Does

• Runtime governance: replay() wraps provider clients and observes tool-call behavior in real traffic.
• Studio review: Governance Studio shows inferred drafts, open judgments, impact preview, approval preview, and conformance findings.
• Provider-agnostic contracts: the same contract model applies across OpenAI, Anthropic, Gemini, and local providers.
• Fail-closed enforcement: production governance fails closed when approved authority is unavailable or invalid.
• Failure classification: failures get stable categories and fingerprints for repeatable investigation.
• Replay evidence: requests, responses, tool definitions, and decisions can be replayed and compared.

Runtime Behavior

After approval, zero-config governance uses the runtime environment to choose behavior:

| Environment | Behavior | | --- | --- | | development | non-blocking monitor behavior | | staging | advisory compatibility evaluation | | production | protective fail-closed enforcement |

Set the environment explicitly or let the SDK derive it:

const session = replay(client, {
  apiKey: process.env.VESANOR_API_KEY,
  agent: "orders-bot",
  environment: "production",
});

Configuration

Common SDK settings:

| Option | Purpose | | --- | --- | | apiKey | Vesanor API key. Usually process.env.VESANOR_API_KEY. | | agent | Stable name for the wrapped agent or workflow. | | environment | Runtime target such as development, staging, or production. | | runtimeUrl | Optional override for the hosted runtime base URL. |

Provider credentials are separate from Vesanor credentials. Configure your provider SDK the way you normally would:

export VESANOR_API_KEY="vsn_..."
export OPENAI_API_KEY="sk-..."
# or
export ANTHROPIC_API_KEY="sk-ant-..."

Exports

The public package exports the runtime SDK from the package root:

import {
  replay,
  observe,
  validate,
  MemoryStore,
  ReplayContractError,
  ReplayKillError,
  ReplayConfigError,
} from "vesanor";

CommonJS works too:

const { replay } = require("vesanor");

CLI

The same package also includes the vesanor CLI for local packs, recorded-provider tests, sync, doctor checks, and replay workflows.

npx vesanor init
npx vesanor

To run a local contract pack against a live provider:

VESANOR_PROVIDER_KEY="sk-..." npx vesanor --provider openai --model gpt-4o-mini

The CLI/local-pack lane is useful for deterministic contract testing and CI. The default hosted product flow starts with replay() and Governance Studio.

Documentation

• Quickstart
• Replay quickstart
• Runtime API
• Governance Studio
• Configuration Matrix
• Security

License

ISC