Sign in

Settings

Signed in

My Profile Name, position, contact, visibility

Security Approval PIN for risky agent actions

Agent action policy Which tools require approval (admin)

Custom Apps Embed external systems by URL

Browser sessions Max concurrent browsers (admin)

Voice Speed Playback speed

Display Font, theme, scale

LLM Provider AI model, API key

Auth Credentials Anthropic OAuth profiles

Default Models Chat, heartbeat

Live Conversation Voice mode, model

Google Accounts Email, Calendar, Drive

Chat Channels WhatsApp, Telegram

OpenRouter Multi-provider API gateway

Business Entities Personal & business profiles

Integrations & MCP servers Connect Pipedrive, Slack & remote MCP — browse the directory

System Timezone, version info

Release Notes What shipped, by version + date

MCP access tokens Expose Wilma's tools to claude.ai / ChatGPT (connector URLs + tokens)

version —

Your name, position, email and team are visible to teammates across the company. Phone and social handles stay hidden unless you turn them on below.

Name

Position / role

What you work on (your assistant remembers this)

Private to you — used only to tailor how your assistant helps you.

Email (your login)

Phone

Show my phone to teammates

Social handles

Show my social handles to teammates

Voice ID

Record a short sample of your voice so meeting transcripts label you by name automatically. Speak naturally for about 20–30 seconds, alone, somewhere quiet. Your voiceprint is biometric data, stored only to recognise you in your team's meetings — you can delete it any time.

Checking…

0:00

I consent to storing a voiceprint of my voice to identify me in meeting recordings.

When Wilma tries a higher-risk action on your behalf (deleting data, payments, sending through external systems), it pauses and asks you to approve it. Set a 4–8 digit approval PIN and the riskiest of those actions will also require the PIN — so a confused or hijacked agent can never push one through on its own. The agent never sees or types this PIN.

Loading…

Set approval PIN

Add an external system by its URL — it opens in a resizable panel inside the app, visible to everyone. Tip: some sites (banks, Google, etc.) block embedding for security; those will show an "Open ↗" link instead of loading inline.

Add an app

1.2x

Voice engine

Voice

TTS Content Filtering

Filter technical text before audio synthesis

Visual chat text stays unchanged. Only spoken audio text is filtered.

Long URLs

Task IDs

Code blocks

Long tokens and hashes

Long file paths

Email addresses

Appearance

Navigation (prototype)

Font Family

Font Size

Document Preview

Panel Text Scale

A A 100%

Token Usage

Show per-message token usage & cost

Shows input/output tokens and simulated API cost under each assistant message

Authentication Method

Provider

API Key

⚫ Not configured

Claude Max Subscription

Use your Claude Max/Pro subscription ($20-180/mo) instead of API keys. This uses the Claude CLI OAuth flow to authenticate.

Setup Steps:

1. Subscribe to Claude Max or Pro

2. Ask your assistant to run the OAuth login flow

3. Complete the browser authorization

⚠️ This method uses your personal Claude subscription for API-like access via the CLI OAuth flow. Works great for personal/team use, but is not officially supported by Anthropic for production SaaS.

Checking...

Paste the code from Anthropic

After authenticating in the browser tab, copy the code shown on the Anthropic page and paste it below.

Manage Anthropic OAuth profiles from UI. Use this instead of CLI setup.

Checking profiles...

Add profile

Profile IDs use letters, numbers, and `:._-`.

Paste the OAuth code from Anthropic

After authenticating in the browser tab, copy the code shown on the Anthropic page and paste it below.

Choose which models to use by default. Saving will restart the gateway to apply changes.

New Chat Sessions

Model used when starting a new chat

Heartbeat Tasks

Model used for periodic heartbeat checks

Manually run a heartbeat check for testing

Configure the real-time voice conversation with Wilma. Changes apply on next connection.

Voice Mode

Direct: Faster responses with streaming. Uses Anthropic API directly with registered tools.
Full Wilma: Routes through the gateway for full Wilma identity, all skills & workflows. Slightly higher latency.

Model

Model used in Direct mode. Full Wilma mode uses the gateway's configured model.

Voice (TTS)

Deepgram Aura 2

Cheapest — uses existing Deepgram balance. Male voice (Orion), English only.

ElevenLabs Flash v2.5

Fast, 75ms latency, 32 languages. Half the cost of Turbo.

ElevenLabs Turbo v2.5 (default)

Best quality, 250ms latency, 32 languages. Used until now.

Cartesia Sonic

Ultra-low latency (40-90ms), ~1/5 cost of ElevenLabs. Multi-language.

Response Style

Brief

1 sentence

Concise

1-2 sentences

Normal

2-4 sentences

Controls how verbose Wilma's voice responses are. Shorter = faster replies. Only affects voice, not text chat.

API Keys

Deepgram

not set

Used for speech-to-text and Deepgram TTS. Get a key at console.deepgram.com

Balance —

ElevenLabs

not set

Used for ElevenLabs TTS voices. Get a key at elevenlabs.io

Characters remaining —

— —

Cartesia

not set

Ultra-low latency Sonic TTS. Get a key at play.cartesia.ai

Status —

Google AI

not set

Required for Gemini models in Direct mode. Get a key at aistudio.google.com

Deepgram Voice

Cartesia Voice

Voice Speed

1.0x

Slower Faster

Response Delay

1200ms

Fast (may cut off) Patient (more delay)

How long Wilma waits after you stop speaking before responding. Lower = faster but may interrupt you. Default: 1200ms.

Silero VAD

Extra voice activity detection. More accurate turn boundaries but adds latency.

Manage your Google accounts and choose which services each account provides.

1 Account Label

A short name to identify this account (e.g. "personal", "work")

Other email providers

Connect a Microsoft 365 mailbox (email + calendar) or any IMAP/SMTP mailbox. These work alongside Gmail in the same email app and for the agent.

Loading…

Connect Google Calendar accounts

1 Choose Account

Account Label

This label identifies the calendar account internally

2 Waiting for Authorization

Complete authorization in the Google tab

Enable Google Drive access per account. Accounts need to be re-authorized after enabling to grant Drive permissions.

After enabling, re-authorize the account in Calendars settings to grant Drive permissions.

Checking...

Enter this code in WhatsApp:

WhatsApp → Settings → Linked Devices → Link a Device → Link with phone number instead

Waiting for confirmation...

Scan this QR code with WhatsApp:

WhatsApp → Settings → Linked Devices → Link a Device → Scan QR

Waiting for scan...

Monitoring

Wilma can observe group chats and notify you about important messages — without responding.

Groups

💡 To find a group ID: forward a message from that group to Wilma on WhatsApp. It'll tell you the group ID.

Contacts

Checking...

Get a token from @BotFather on Telegram

Chat channels let Wilma receive and send messages on WhatsApp and Telegram. Configure access policies and allowlists in the gateway config.

Access 300+ AI models (GPT-5, Llama 4, Mistral, etc.) through OpenRouter's unified API. Get an API key

API Key

Connect an app

Pick an app and paste one API key — no URL or transport setup. Its tools become available to the chat agent (reads are on by default; turn on Writes per app to allow changes). Test checks the key regardless of the enable toggle.

Connected

Garmin Connect

Checking...

OAuth connections (admin)

External MCP servers Wilma connects to

Register an external MCP server (URL + bearer token) so Wilma can query it. Its tools become available to Agents & Automations as remote: steps. Turn on Agent access to also let the chat agent use that server's read tools — register a read-only-scoped token for that.

Curated directory

Popular MCP servers + first-party integrations, curated centrally. Pick one — we pre-fill the connect form so you only add a token (or click Connect for OAuth).

AI assistant name

The name your assistant calls itself, shown across the app and used in its system prompt. Saving applies it for everyone immediately — all open sessions reload.

Timezone

Current: —

Timezone saved. Container restart required for full effect.

Version

OpenClaw—

Node.js—

Connector URLs

Add any of these as a custom MCP connector in claude.ai, Claude Desktop, ChatGPT, or Claude Code. Each authenticates on first connect.

Access tokens

Bearer tokens for clients that can't use the connect-and-authenticate flow. Each token is named per surface (e.g. claude-ai-desktop) and can be revoked here at any time.

Wilma

Settings

Add MCP server

New MCP Token

Token created — copy now

Recent calls —

Threads

Apps

Import Contacts / Companies

Email Campaigns

New Campaign

Select Recipients

Campaign Summary

⬛ Terminal

🤖 Oscar Sessions

Your workspace

Welcome back

Reset your password

Wilma

📹 Send Wilma to a meeting

Add MCP server

New MCP Token

Token created — copy now

Recent calls —

Threads

Add Company

Import Contacts / Companies

Email Campaigns

New Campaign

Select Recipients

Campaign Summary

⬛ Terminal

🤖 Oscar Sessions

CloudMax Usage