Product roadmap

Email sign-in for operators. Session cookies gate projects and the upload UI.

Create a VoiceThere account from /signup with email magic link or GitHub, Google, or LinkedIn — no waitlist.

Sing in via GitHub, Google, or LinkedIn.

Create projects, upload verified agent bundles, and inspect build history from the dashboard.

Set which build is active for a project or roll back to the previous release.

REST at /api/v1 for projects, builds, promote, and rollback — same contract as the CLI.

Spec-compliant /llms.txt and /llms-full.txt listing deployed features, on-platform STT and TTS, plan prices, docs, OpenAPI, sessions, webhooks, and npm packages.

Published @voicethere/cli — login, build upload, deploy --wait, and promote. Pair with @voicethere/agent verify in CI.

Roll out your active build from the dashboard or CLI — voicethere deploy --wait promotes and provisions managed voice workers. A failed deploy says whether to fix your agent or wait for us, and offers a ticket when the cause is on our side.

Start a session from your backend; receive a join token and signaling URL for browser WebRTC. Async provisioning when workers scale up.

Test voice and text against echo agents from the project page, or paste an embed snippet with a client API key.

Start a voice session even when the microphone is denied or unavailable, switch input devices from the dashboard chat, and see an in-chat notice when permission is blocked (agent audio can still play).

Invite teammates, assign org roles, and restrict projects to specific members from the dashboard.

Immutable trail of API and dashboard actions — browse org-wide or per project.

A Metrics area on each project with Overview, Sessions, Voice and Agent tabs: success rate, time to connect, concurrency against your plan limit, failures by reason, and agent errors. Credits and billable time have their own Usage page. Click a failure reason to open exactly the sessions behind it, already filtered.

Voice response time, finalize delay, STT and TTS latency by vendor and language, abandoned turns, and barge-ins on the Metrics Voice tab and API, with only your project's data.

Printable credit breakdown by project and usage kind, plus links to Stripe hosted invoices for subscription and metered overage cash charges.

Flat monthly plans via Stripe Checkout and Customer Portal. Subscribe, assign subscriptions to projects, and buy multiple slots of the same tier.

Each tier includes monthly usage credits that refresh on your subscription billing cycle. Credits cover voice, data, speech, and other metered usage; new sessions are gated on available credits per project.

Paginated session list per project with status, duration, and billable seconds — same records as CLI sessions list and billing queries.

Several clients in one session with shared or per-peer agents. Disable voice for DataChannel-only apps — games, live sync, and chat without STT/TTS.

Configure idle timeouts, data-only mode, shared vs per-peer agents, and cloud voice providers from the dashboard or CLI.

Pick OpenAI, Deepgram, ElevenLabs, Fish Audio, Google, Groq, Azure Speech, AWS, or VoiceThere's own speech per project — with documented streaming models, secrets, and customer-paid vendor usage.

Structured errors from agents and workers in the dashboard — filter by session and map codes using the public error catalog.

Write structured agent logs and search them in the dashboard — filter by message, level, and fields per project or session.

Searchable structured logs from agentLog and console.* — filter by message, level, and JSON fields. Session failures persist as error-level lines with stable codes and soft rollover. Paid overage past the monthly included cap bills usage credits (1 per 1,000 entries) when overage billing is on.

Transcripts, agent replies and your agent's error messages are written only to your project's agent logs, never to VoiceThere's internal logs.

Developer documentation at /docs — quickstarts, API reference, session guides, agent environment, secrets, deploy flow, and package install instructions.

Search developer documentation by keyword and browse guides grouped under Guides and Help on /docs.

Try voice, multiplayer sync, and spatial audio from the marketing site — landing chat, shared world, orbiting voice, spatial soundboard, and proximity room.

Product blog and email newsletter at /blog — release notes, guides, npm updates, dashboard features, and platform news.

Starter templates include matching client-side UI alongside the agent bundle. Hosted client assets are served from a global CDN and cache-purged when you deploy a new build so browsers pick up the latest embed or demo code.

Advanced and Ultimate projects get a private Redis instance for your agents — credentials injected at start. Sync game state, lobbies, and context across every concurrent session worker in the same project.

POST HTTPS to your project URL — the platform fans requests to every ready session worker. Verify signatures in your agent; warm pool keeps a replica ready between sessions.

Fine-tune STT/TTS, barge-in, inbound RNNoise (on by default for voice runners), and related voice behavior from the dashboard or API — beyond basic provider selection.

A call keeps its session through a network drop of about 10 seconds, or about 25 seconds when you raise the project's reconnect window to 30 s. If the caller never comes back, billing stops at the drop. Needs @voicethere/client 0.11.9 or later.

Deploying a new build or scaling down no longer cuts off calls in progress. They keep running on the previous version for up to 9 minutes while new calls start on the new build. For an urgent fix, deploy with Force instead: it ends calls in progress right away so every new call gets the new build.

Sidebar navigation, live project status in the header, a setup checklist for new projects, and a dedicated project Settings tab.

Replies your agent says often play from memory without being synthesized again, and replays don't use speech credits. Replies that start a new sentence after a period also begin sooner. Both are on by default, with project settings to turn them off.

Agents keep listening and can be interrupted on time even when a transcription is slow. Busy servers no longer drop words from calls, and a short first word like "Hi" stays in the transcript.

The project Usage page shows characters, credits and sessions for each voice, and splits usage exactly when a call changes voice mid-session.

Measured median and p90 latency for VoiceThere speech at 1, 10 and 45 concurrent conversations, and for ElevenLabs, Cartesia, Groq, OpenAI and Fish Audio. The page lists the method and the commands to repeat the runs on your own account.

Detect the caller's language and switch speech-to-text and voice during the call, automatically or from agent code, without dropping the connection. While a new language loads, callers hear a short message naming it. If it can't load, the call continues in the original language and your agent gets a language_unavailable event.

Component health for voice and related services — operators and customers can check platform status.

Relay servers included in session mint responses — reliable calls through firewalls and NAT.

Ensure that every paid tier has guaranteed resources in the cluster, free tier can be queued or not available.

Set max concurrent sessions by STT/TTS choice — local/self-hosted voice uses more compute but costs less, so subscription tiers expose different concurrency caps than cloud vendors.

Browse past STT and TTS turns from the dashboard (encrypted at rest, with PII masked where possible). Your agent can turn transcript storage off for one session while collecting a PIN or card number. DataChannel chat is not stored. Each turn has a details panel with the model or voice, timings and an estimated credit cost.

Opt-in voice-session audio with dashboard playback, included minutes per plan, rollover bank or metered overage, and agent start/pause/resume/stop for sensitive moments.

Dynamic audio mix groups within a session — reassign who hears whom mid-call, mute clients globally or per listener, and inspect mix status with getClientMixStatus.

Live stereo poses with setTtsPose and setPlayPose — the orbit showcase loops a sine tone and TTS around each listener, with mute controls and per-client STT overrides on Voice+Data mix projects.

Play sound effects and notification clips to one or all listeners — HTTPS URLs or small inline bytes, with play-id progress and stop control. Clips mix with TTS and microphones but are not speech synthesis.

Edit agent TypeScript in the dashboard Code tab (file tree and Save), download saved source as a zip or compiled agent.js from the Code tab and Overview build history, then Compile & upload, Promote this build, and Deploy to cloud as three separate actions. Seeded templates include a root package.json with @voicethere/agent and npm scripts; older workspaces can add one from the tab. Refresh types updates dependency IntelliSense. Public npm semver packages are always available, with per-plan caps on stored source size and the editor typings pack.

Complete quickstart, API reference, browser client guide, and deploy runbooks at /docs.

Register HTTPS webhook URLs that VoiceThere delivers into your running session workers — payments, CRM events, or custom integrations reach the same process that owns the voice call.

Dedicated Widget tab with live preview chat, eight launcher positions plus custom offset, fonts and chat bubbles, optional custom CSS — Deploy publishes appearance to the VoiceThere CDN so customer sites pick up theme changes without editing their page.

Enable or disable speech-to-text from agent code on voice+data sessions — turn recognition on for meeting summaries and space notes, or off when you only need media and data.

Optional widget setting streams agent speech into the chat bubble word by word while audio plays — enable from the Widget tab or streamSpokenText in published config.