Insights Chat
Configure your database to get started
Hello! How can I help you explore your data today?
Ask questions in English, Spanish, Hindi, Turkish or any language — I'll query your database and explain the results.
⌘K to focus · Shift+Enter for new line
MCP Agent
Multi-agent data processing & orchestration
MCP Agent Ready
Powered by multi-tool orchestration. The agent can chain SQL queries and external tools for complex tasks.
⌘K to focus · Shift+Enter for new line
Documents
Upload files. We'll flag any sensitive content automatically.
No documents yet
Upload a document above to get started.
| Name | Flags | Size | Uploaded | Status |
|---|
Manage flag categories
Define what content to flag in your uploads. Each user's list is private. Disable to keep history but stop flagging new uploads.
Golden Rules
Persistent rules that shape how AI generates SQL and answers
- You can also add rules from chat: say "Golden rule: always exclude test accounts"
- Rules are injected into every query. Keep them concise and specific.
- Toggle rules off without deleting to temporarily disable them.
- SQL rules affect query generation. Answer rules affect how results are formatted.
Proactive Insights
AI-discovered patterns, anomalies, and correlations
No insights yet
Click "Scan Now" to have AI analyze your data sources for patterns, anomalies, and correlations.
Active Monitors
Usage
Loading your usage…
Loading your usage…
No usage yet this period
Your billed usage will appear here once you start using the chat.
| Date | Chats | Tokens | Billed |
|---|
Billing period: current calendar month (ET). Amounts include Arcturus platform fees.
Admin
Fleet-wide usage, per-user controls, pricing, and configuration.
Loading admin data…
| Model | Raw | Billed | Margin | Calls | Tokens | Last seen |
|---|
| Request type | Calls | Raw |
|---|
| Provider / Model | Calls | Tokens | Raw |
|---|
Override per-user model assignment, markup %, content storage, eval credit, and billing status. Changes apply on the user's next request.
| Assigned model | Effective model | Markup % | Store content | Credit $ | Used $ | Status | Last seen |
|---|
Per-1M-token rates. Editing an active row supersedes it (historical usage rows keep their original pricing_id).
| Provider | Model | Input $/1M | Output $/1M | Effective from | Status |
|---|
Admins see every customer's raw cost, can reassign models, and edit fleet config. The bootstrap admin cannot be removed.
| Added by | Added at |
|---|
Fleet-wide defaults. Per-user overrides on the Users tab take priority over these.
Data Sources
Configure one or more database connections
- Ensure the DB user has SELECT privileges on the listed tables.
- List only tables relevant to your queries for best AI accuracy.
- Use Test Connection before saving to verify credentials.
- For cloud databases, whitelist your server's IP address first.
- Table names are case-sensitive on Linux MySQL servers.
- Databricks: Use a Personal Access Token as the password. Set username to
token. Find the HTTP Path in your SQL Warehouse → Connection Details tab.
AI Provider
Choose your language model backend
Get your key at platform.openai.com
- gpt-4o — Best overall, multimodal
- gpt-4-turbo — Faster, cost-effective
- gpt-3.5-turbo — Fastest, cheapest
Get your key at console.anthropic.com
- claude-sonnet-4-6 — Latest, most capable
- claude-3-5-sonnet-20241022 — Excellent balance
- claude-3-haiku-20240307 — Fast & affordable
Get your key at aistudio.google.com
- gemini-1.5-pro — Most capable
- gemini-1.5-flash — Fast & efficient
- gemini-pro — Standard model
MCP Configuration
Configure MCP servers and agent behavior
MCP Agent is a multi-step agent that combines your data (via SQL) with external tools (via MCP servers) to complete tasks a single SQL query couldn't.
- Insights Chat answers questions about your data — one SQL query, one answer.
- MCP Agent can chain multiple steps: query data, transform it, then act on it (send a message, create a ticket, write to a sheet).
- Add an MCP server above (Slack, Gmail, Google Drive, GitHub, etc.). Paste the server URL and auth token.
- Click Test Connection. A green check means the agent can reach it.
- Open MCP Agent in the sidebar and ask a multi-step question.
- "Find our top 10 customers this quarter and email the list to alex@acme.com." (SQL → Gmail)
- "Pull yesterday's failed orders and post a summary to the #ops Slack channel." (SQL → Slack)
- "Open a GitHub issue for every product with stock under 5." (SQL → GitHub)
- Agent Timeout — total seconds before the agent gives up. Increase for long workflows (default 60s).
- Max Tool Calls — safety cap on how many tools the agent can chain in one turn (default 10).
- Show tool traces — display each step the agent takes inline. Useful for debugging.
Settings
Notification and system preferences
By default ArcIn only answers questions about your selected data sources. Turn this on to let it also answer general questions (definitions, coding help, explanations) when the question doesn't fit your data.
Used as a fallback for real-time questions (weather, news, current events). Powered by Tavily. Free tier ≈ 1k queries/month. Add your own key here to bypass shared usage limits.
A small classifier picks the right path per question — your databases and documents stay primary; general fallback only fires when nothing else fits.
A small agent you run inside your network so ArcIn can query databases behind a firewall. Database credentials stay local — only query results travel back over an outbound, encrypted WebSocket.
- Extract
arcin-connector.zipon a machine inside your network that can reach the database. - Open
README.txtfor platform-specific install commands:- Linux:
sudo ./install-linux.sh(systemd service) - macOS:
./arcin-agent --config agent.yaml - Windows: right-click
install-windows.ps1→ Run with PowerShell (admin)
- Linux:
- Edit
agent.yamlwith your database DSN (stays local). - Start the connector. It prints a Customer ID and Source ID — use those when you add a connector-mode source under Insights Chat → + Sources.
To upgrade an existing install: stop the service, replace the binary with the one from the new zip, restart.
Configure push alerts for monitor triggers. Uses ntfy — self-hosted or public.
- Public ntfy.sh — free, no auth needed. Topic names are public, so pick something unique.
- Self-hosted — run
docker run -p 2586:80 binwiederhier/ntfy serveand set the URL to your server. - Install the ntfy app on your phone (iOS/Android) and subscribe to your topic to receive alerts.
- Monitor alerts fire when a watched metric changes by more than its threshold %.