Namespaces

Beyond chat, streaming, realtime, and RAG, the client exposes several more namespaces. Each is lazily initialized — accessing client.user, for example, constructs that namespace on first use.

User

client.user — the signed-in user’s account surface.

Sub-namespace	Methods	Purpose
user.balance	get()	Current wallet balance.
user.payment	listMethods(), deposit(body), getTransactions(range?)	Payment methods and history.
user.preferences.audio	get(), update(prefs)	Audio tool preferences.
user.savedReleases	get(), add(releaseId), remove(releaseId)	Bookmarked Releases.

const balance = await client.user.balance.get();
await client.user.savedReleases.add("rel_abc123");

API keys

client.apiKeys — manage workspace-scoped API keys from the client.

create(params, body), list(params?), get(params), update(params, body), delete(params), rotate(params), listUsers(params).

const created = await client.apiKeys.create({ workspaceId: "ws_123" }, { name: "CI key" });

BYOK (bring your own key)

client.byok — register and rotate your own model-provider credentials.

create(params, body), list(params?), get(params), updateMeta(params, body), rotate(params, body), delete(params).

Audio

client.audio — upload and manage audio transcripts, and turn them into RAG or fine-tune material.

Top-level: upload, uploadFromUrl, list, get, update, generateRagFile, generateFineTuneFile, delete. Plus audio.voiceprints and audio.bulk sub-namespaces.

const transcript = await client.audio.uploadFromUrl(
  { workspaceId: "ws_123" },
  { url: "https://example.com/call.mp3" },
);

Arena

client.arena — model A/B comparison. Enable/disable, fetch results, select chat/transcript variants, regenerate a variant, run tests, and estimate cost. Presets live under client.arena.presets. See the server Arena page for the full conceptual model.

Fine-tune

client.fineTune — create and manage fine-tunes and their training files from the client. create, list, get, update, fork, delete, plus fineTune.files (upload, list, get, getContents, delete).

Style patterns

client.stylePattern — manage tone/style rewrite rules: list active and proposed rules, test a rewrite, approve/reject proposals, toggle rules, and read the audit log.

Notifications

client.notifications — per-workspace notification feed.

list, get, markAsRead, setTags, getCounts, getTags, plus sub-namespaces: notes(...) for annotation, channels(...) for email/webhook delivery, triggers(...) for tag-routing rules, and flaggers(...) for custom LLM-driven notifications.

const counts = await client.notifications.getCounts({ workspaceId: "ws_123" });

See Notifications, Analytics & Metrics for the full custom-notification pipeline (flagger → trigger → channel), product analytics, and pipeline-metrics alerting.

Analytics & Metrics

client.analytics — product analytics (metrics, trends, experiments, funnel, trackEvents ingestion). client.metrics — pipeline telemetry + custom alert configs. Both documented on the observability page.

Terms of Service

client.terms — check and accept a release’s published Terms of Service (message sends are blocked server-side until the current version is accepted):

Namespace	Methods	Purpose
terms	getForRelease({ releaseId? }), accept({ tosId, version, releaseId? })	Fetch the ToS gate for a release; record acceptance for the authenticated identity.

const gate = await divinci.terms.getForRelease({ releaseId });
if (gate.required) {
  // present gate.title / gate.content to the user, then:
  await divinci.terms.accept({ tosId: gate.tosId!, version: gate.version! });
}

Feedback

client.feedback — the operator’s read side of message feedback (consumers submit it via client.chat.submitFeedback): list(filters?), stats(), get(params), plus withWorkspace(workspaceId) to bind a workspace. Documented in full on the observability page.

Site control

client.siteControl — execute the navigation/action directives a deployed assistant emits (divinci-action blocks), same-origin. Fetch action descriptors with loadDescriptors(releaseId) (or setDescriptors(...) from freeChatGate.getConfig().siteActions), register handlers with registerHandler(name, handler), then run each assistant reply through process(reply) to strip the directive block and perform the action. Helpers: prepare(reply), setConfirm(fn), setAutoSubmit(enabled), setNavigator(fn). See Assistant Tools for the server-side siteControl release config that gates this.

await divinci.siteControl.loadDescriptors(releaseId);
const { reply } = await divinci.freeChatGate.send("take me to pricing");
const { cleanedText } = await divinci.siteControl.process(reply); // navigates

Public / gated chat

Three namespaces power unauthenticated chat experiences:

• client.anonymousChat — generic anonymous chat against a release with allowAnonymousChat: true — no account, no email gate. send(prompt), sendDelegated(...) (on-device inference), getTranscript(), createHandoff(opts?), hydrate(...), getState(), reset(). This is the public release-backed chat used by embed / docs-assistant widgets.
• client.homepageChat — Turnstile-gated email verification then send: verifyEmail(email, turnstileToken), confirmEmail(email, code), send(prompt).
• client.freeChatGate — the configurable free-chat gate (captcha / OTP / magic-link) with language support: start(args), verifyOtp(args), state(), send(prompt), submitFeedback(...), plus token and transcript helpers.

These back the Embed script and Divinci’s own landing pages.