Chat with your rules through the OpenAI or Anthropic API

Ask a deployment questions in plain language from any OpenAI or Anthropic client. Legalese Cloud serves each deployment in both formats, backed by the same legalese-comply-4 model pipeline: the model decides which exported rules to run, gathers their inputs and explains the result. It is the same service the Use in chat button in the VS Code Deployments tab connects to.

Before you start: a deployment on Legalese Cloud (Deploy your rules). To compare this with the other integrations, see Choose an integration.

---

Endpoint

https://ai.legalese.cloud/{orgSlug}/{deploymentId}/v1

{orgSlug} is your Legalese Cloud organization; {deploymentId} is the deployment's name. The VS Code Integrate dialog pre-fills both for you. Append the path that matches your client's wire format:

Client wire format	Append
OpenAI	/chat/completions
Anthropic Messages	/messages

Both paths lead to the same pipeline, with the same authentication and daily token quota; only the shape of requests and responses differs. Use whichever SDK your stack already has.

Authentication

Send your credential, a Legalese Cloud session or an sk_... API key, in either header:

• Authorization: Bearer {token}: the OpenAI SDK's default, and what the VS Code extension sends with your session;
• x-api-key: {token}: the Anthropic SDK's default.

An API key needs the ai:chat, l4:rules and l4:evaluate permissions. With ai:chat but not l4:evaluate, the model can hold a conversation but cannot run any rule, and says it cannot compute the answer. See API keys and permissions.

Use it

OpenAI clients

curl

curl https://ai.legalese.cloud/{orgSlug}/{deploymentId}/v1/chat/completions \
  -H "Authorization: Bearer sk_..." \
  -H "Content-Type: application/json" \
  -d '{
    "model": "legalese-comply-4",
    "messages": [
      { "role": "user", "content": "Is a tenant who is 40 days late on rent in breach?" }
    ]
  }'

OpenAI SDK (Python)

from openai import OpenAI

client = OpenAI(
    base_url="https://ai.legalese.cloud/{orgSlug}/{deploymentId}/v1",
    api_key="sk_...",
)

resp = client.chat.completions.create(
    model="legalese-comply-4",
    messages=[{"role": "user", "content": "Is a tenant 40 days late in breach?"}],
)
print(resp.choices[0].message.content)

OpenAI SDK (TypeScript)

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://ai.legalese.cloud/{orgSlug}/{deploymentId}/v1",
  apiKey: "sk_...",
});

const resp = await client.chat.completions.create({
  model: "legalese-comply-4",
  messages: [{ role: "user", content: "Is a tenant 40 days late in breach?" }],
});
console.log(resp.choices[0].message.content);

Streaming (stream: true) and multi-turn conversations work exactly as they do against api.openai.com.

Anthropic clients

curl

curl https://ai.legalese.cloud/{orgSlug}/{deploymentId}/v1/messages \
  -H "x-api-key: sk_..." \
  -H "anthropic-version: 2023-06-01" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "legalese-comply-4",
    "max_tokens": 1024,
    "messages": [
      { "role": "user", "content": "Is a tenant who is 40 days late on rent in breach?" }
    ]
  }'

Anthropic SDK (Python)

import anthropic

# Anthropic's SDK appends "/v1/messages" itself, so base_url stops at the deployment root.
client = anthropic.Anthropic(
    base_url="https://ai.legalese.cloud/{orgSlug}/{deploymentId}",
    api_key="sk_...",
)

resp = client.messages.create(
    model="legalese-comply-4",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Is a tenant 40 days late in breach?"}],
)
print(resp.content[0].text)

Anthropic SDK (TypeScript)

import Anthropic from "@anthropic-ai/sdk";

// baseURL stops at the deployment root; the SDK appends "/v1/messages".
const client = new Anthropic({
  baseURL: "https://ai.legalese.cloud/{orgSlug}/{deploymentId}",
  apiKey: "sk_...",
});

const resp = await client.messages.create({
  model: "legalese-comply-4",
  max_tokens: 1024,
  messages: [{ role: "user", content: "Is a tenant 40 days late in breach?" }],
});
console.log(resp.content[0].type === "text" ? resp.content[0].text : "");

Streaming (stream: true), extended thinking, and multi-turn conversations work exactly as they do against api.anthropic.com. The model's reasoning arrives as ordinary thinking content blocks.

Notes

• The endpoint knows only the rules exported in that deployment. It has no access to your editor, your workspace or your other deployments.
• The model name is always legalese-comply-4. Both formats reject other model names, such as claude-opus-4-7, with 400 model_not_found.
• A self-hosted jl4-service does not offer this API; use MCP or REST there.