Connect to your providers to get started. Agents need at least one AI provider before they can chat.
Welcome!
No AI providers are configured yet. Ask your administrator to add a provider before creating an agent.
Welcome!
Create your first agent to get started. Give it a name, description, and role.
Working on it
Agents in this chat
When this chat delegates work, agents and their progress will appear here.
Original chat
Add Widget
0 selected
Skills
Choose a skill to open its drawer.
Agents
Inbox
Loading…
Email
News
Loading…
Markets
Loading market data…
Today
Loading…
Needs attention
Check approvals, interrupted work, and connection problems across your account.
Tasks & Ledger
Routines
Save an action sequence and run it again with a phrase.
No tasks yet.
Start with "remind me" for a chime + banner only.
Anything else runs the agent — results appear in the Ledger tab. To get pushed,
include "and send me a notification when done" or "text me the result".
Learn
Loading…
Run Inspector
No runs recorded yet.
Memory
Loading memories…
Skill Permissions
Loading permissions…
Documents
📤
Click to upload a document
PDF, images, text files — up to 150 MB
📄
SHARE WITH
0 selected
Share
SHARE WITH
Messages
New Message
To:
Expenses
Overview
This Month
—
This Year
—
All Transactions
—
Top Category
—
Spending by Category
Total
—
Recent Transactions
Date
Merchant
Category
Amount
Transactions
Date
Merchant
Description
Category
Amount
No transactions found — try adjusting your filters or import a statement.
Reports
Import Statement
Drop files here or click to browse
Receipts, invoices, bank statements — PDF, JPG, PNG, CSV
Extracted Transactions
Date
Merchant
Category
Amount
Shared Group
Loading…
Create Shared Group
Select members to share expenses with:
Add member:
Settings
Network Shares
Connect SMB or NFS folders to your account. Browse them here, or ask in chat: “Connect to the shared folder on shareserver” and “Copy my report to that share.”
Find shared folders on a server
Connect a share
Leave login fields empty to use a saved login or guest access.
NFS uses server export permissions and the OE service identity. It does not use an SMB username or password.
Agent setup
Loading your setup…
Assign a model to each agent
Configure LLM provider endpoints and API keys. Toggle providers on/off to control which models agents can use.
Provider updates
OE checks for updates every six hours. Supported changes are tested when you use the provider, then saved after a successful response. Failed tests keep your previous settings. Test requests may use a small amount of your provider allowance.
Loading update status…
Anthropic
API key for Claude models.
Fireworks AI
Fireworks has deprecated image generation. Only image models still returned by its API are listed. Choose another image provider if your model is unavailable.
Ollama (cloud)
Use your Ollama Cloud account with an API key. OE checks the key before saving it.
Ollama (local)
Base URL for a locally-installed Ollama instance.
xAI Grok
API key for xAI Grok — chat (grok-4, grok-3, grok-3-mini), image (grok-imagine), and video (grok-video) models.
LM Studio
Base URL for your local LM Studio server.
OpenRouter
API key for OpenRouter — access hundreds of AI models through a single endpoint.
Text-to-Speech
Pick the TTS provider. Local options (Piper, KittenTTS, Pocket TTS) run via systemd services on this box; remote options need an API key.
0.85×
slowerdefault 1.0faster
Piper runs locally — no API key required. Checking service…
More voices can be downloaded from the catalog after install — no service restart needed.
1.10×
fasterdefault 1.0slower
Voice catalog — downloaded voices appear automatically and become selectable per slot in Voice Devices.
KittenTTS is a tiny CPU-only TTS engine (no GPU, no API key). 8 preset voices, no voice cloning. Checking service…
Pocket TTS is a tiny CPU-only engine with zero-shot voice cloning — no GPU, no API key. Checking service…
Tip: use a clean ~15–20 second clip of a single speaker. Voice synthesis by
Kyutai Pocket TTS,
licensed CC-BY-4.0.
Speech-to-Text
Required for voice devices to transcribe captured speech.
OpenAI-compatible /audio/transcriptions endpoint.
Checking service…
Removed only after the new model is ready. Downloaded models are kept for faster reinstallation.
Only the selected local speech service runs. Switching stops the previous service and frees its memory. Both Photon choices include Silero to reject recordings with no detected speech.
Enable or configure drawers. Disabled drawers are hidden from the sidebar.
Loading drawers…
Roles
Roles define what each agent is responsible for. Claim a role to assign it exclusively to one agent.
Loading…
Custom skills
Skills you built or installed. Each one belongs to a single agent — pick which one below, or chat with an agent and say /claim <skill-id>.
Loading…
Tools
Tools are capabilities available to any agent.
Loading…
Reminder Delivery
Where your scheduled reminders get delivered when they fire. "All" sends to every channel you have set up.
Send reminder emails from:
Speak reminders on:
"All channels" speaks on every paired device — the preferred pick above only applies to the standalone "Voice device" channel.
Tasks & Ledger
No tasks yet.
Start with "remind me" for a chime + banner only.
Anything else runs the agent — results appear in the Ledger tab. To get pushed,
include "and send me a notification when done" or "text me the result".
Auto-Label
Tag incoming emails automatically — Gmail accounts only
Control which tools each user can access, lock their settings, or adjust roles.
Chat with your coordinator agent from anywhere via your own Telegram bot. Create one with @BotFather, paste the token, register a public HTTPS webhook, then send /start from your bot.
Loading…
Restart Server
Fully shut down and respawn OpenEnsemble. All in-flight chats and WebSocket connections will be dropped.
Public Access (Cloudflare Tunnel)
Expose this install to the public internet so webhooks (Telegram), OAuth callbacks, and sharing links work without a static IP. Server-wide setting — owner/admin only.
Loading…
Private Mesh (Tailscale)
Join this install to your tailnet for a stable private IP reachable from any device on your Tailscale network — no public port exposure. Server-wide setting — owner/admin only.
Loading…
OpenEnsemble ModelsCortex • Scheduler
Cortex Memory
Models used to embed memories and score importance. Any provider you have enabled is available here.
Scheduler
Fine-tuned model that interprets scheduling requests ("remind me to…", "every morning…"). Picking Ollama or LM Studio pushes our model into that runtime.
Expense Receipt Model
Parses receipts, invoices, and statements uploaded directly through the Expenses drawer. Chat attachments use the agent's own model and ignore this setting.
Loading…
Brave Search API
API key used by the Web and Deep Research skills and the News drawer. Server-wide setting — changes apply to all users. Get a key at api-dashboard.search.brave.com.
Loading…
"Model-native only" turns off billed Brave searches in chat. Models with built-in web search (ChatGPT, Grok, Claude, Perplexity) keep searching; models without it lose web search, and Brave-backed Deep Research stops working. The News drawer always uses Brave (it doesn't go through a model).
Home Assistant
One household connection with live state events. Any user who enables the Home Assistant role on an agent can control your smart home, and HA automations can fire approved OE routines. Create a Long-Lived Access Token in Home Assistant (Profile → Security → Long-Lived Access Tokens).
TV Video Sources
Folders the TV video library serves — local folders on this server or guest-accessible SMB network shares (//host/share). Shared by every TV in the household.
Loading…
Session idle timeout
Strip Thinking Output
Hide internal reasoning blocks (<think>…</think>) emitted by reasoning models before showing their response.
Server Logsrecent events
View recent server events. Logs rotate automatically at ~5 MB.
—
Loading…
Register Model Context Protocol servers and assign them to your agents. Each server runs as a subprocess on this host with the env vars you provide (encrypted at rest); the tools it exposes get merged into the agents you assign. Adult accounts only.
Loading…
External access — use your agents from other apps
This server is also an MCP server. Create an access token and add it to Claude Code, Claude Desktop, Cursor, or any MCP client to talk to your agents and memory from there. Agent tools are read-only by default; additional exact tools must be granted on the token. Anything you send through an external client flows to that client's AI provider and may appear in its logs. Adult accounts only.
Loading…
OpenEnsemble Bridge lets agents request access to tabs you explicitly lease, ask to open URLs, and perform confirmed browser actions. A connection alone exposes no tab list or page content. LAN-only — the extension talks only to your OE server.
Loading…
Download a full backup of your OpenEnsemble data, or restore from a previous backup. Backups include config, users, sessions, expenses, tasks, and autolabel rules.
Export Backup
If you set a password, the backup is encrypted with AES-256 and the password is required to restore. There is no recovery — if you forget it, the backup is unreadable.
Restore Backup
Warning: This will overwrite current data. The server should be restarted after restoring.