Hand all your database
monitoring to AI.
Just register a connection string. AI handles everything from deciding which numbers to watch to finding the cause and proposing a fix, so you catch problems before your customers do.
- No monitors to design, no thresholds to tune
- Alerts arrive with the cause and a proposed fix
- Connects read-only and keeps your customers' personal data away from the AI
ScandockMonitoring
Settings
DB・Orders in the last hour
Incoming order volume. A sudden rise or drop suggests a problem with payments or the purchase flow.
DB・Payment outcomes
Watches successful and failed payments together.
DB・Shipping job status
Breakdown of pending shipping jobs by status.
- Waiting62%
- Running24%
- Retrying14%
DB・Failures by payment method
Compares failures in the last hour by payment method.
# db-alerts
Scandock14:03Payment error rate at 6x its normal range
The cause is slow responses from one card issuer. We proposed changing the retry interval.
Alerts and actions go to the tools you already use
Slack
Teams
Google Chat
LINE WORKS
Chatwork
kintone
Discord
LINE
PostgreSQL
MySQL
MariaDB
SQL Server
TiDB
Cloudflare D1
BigQueryAlso works with Neon, PlanetScale, Azure SQL Database, Amazon RDS and Cloud SQL.
Do these database problems
sound familiar?
You hear about problems from your customers
Orders or payments can stall, and nobody notices until a customer gets in touch. By then, the losses have already piled up.
Everything depends on one or two people
Only the few members who know the database can make sense of it. When they're away, investigation and decisions grind to a halt.
Finding the cause eats up your time
Every alert means tracing queries and logs by hand. Nights and weekends, it's the same people handling it.
Handing over production feels risky
You don't want to grant strong privileges, expose the database, or hand your data to an outside AI.
Scandock takes care of all of them.
Watch, investigate, act.
An AI agent keeps the loop running, around the clock
Scandock's AI agent repeats six stages, from designing the monitoring to taking action. Actions that write run only after a person approves them.
- STEP 01
Design monitoring
Decides what to watch and how often, from the schema and real data
- STEP 02
Collect values
Runs queries on a schedule with a read-only user
- STEP 03
Judge
Judges from trends, past history and your team's knowledge
- STEP 04
Investigate
Follows related data to where the cause lies
- STEP 05
Alert and propose
Sends the cause and a proposed fix to your tools
- STEP 06
Approve and run
Runs a registered action once it is approved
3 things that set Scandock apart
AI takes over designing your monitoring
Just connect, and AI reads your database's structure and real data, then decides which numbers to watch and how often to check them. AI judges anomalies instead of fixed thresholds, so there are no thresholds to set or maintain by hand. As your service changes, AI keeps revising its monitors. It also remembers judgments such as "a rise at this hour is normal", so false alarms decrease the more you use it.
Payment error rate at 6x its normal range
- Cause
- Failures concentrate on slow responses from one card issuer
- Fix
- Temporarily lengthen the retry interval
Every alert comes with the cause and a fix
Each alert already contains the judgment, the results of the investigation, and a proposed fix. What you need to decide is knowledge of your business, not SQL.
Safe to trust with production and customer data
Monitoring and investigation use a read-only user, and writes run only as registered steps after approval. Your customers' names and email addresses are never sent to the AI provider. Databases not exposed to the internet can be reached through Cloudflare Tunnel.
Your customers' personal data never reaches the AI
The AI investigates with names, email addresses and other personal data masked. You keep the quality of the investigation without handing personal data to an outside AI.
What you see
| name | orders | |
|---|---|---|
| taro@example.com | Taro Yamada | 12 |
| hanako@example.jp | Hanako Sato | 3 |
What the AI receives
| name | orders | |
|---|---|---|
| 12 | ||
| 3 |
Never reaches the AI provider
The AI receives your data with customers' names and email addresses masked. Their contents are never sent to the AI provider.
Investigation stays sharp
Even with values masked, it pins down duplicates and skewed counts. Notifications and the console show the original values.
See what was masked
Values masked from the AI are outlined with a dotted border, so you can always check what the AI received.
Designed to be trusted with production
Rather than handing AI privileges, the implementation narrows what AI can touch. Reads and writes go through separate connections, and the only path to a write goes through approval.
Runs as a read-only user
Monitoring and investigation use only a read-only user. If you register a connection that can write, a dedicated user is created on the spot.
Single read-only statements only
Queries generated by AI run only on the read-only connection and are limited to a single SELECT / EXPLAIN / SHOW statement.
Writes follow only registered steps
Actions that write are limited to ones registered in advance. Anything that uses admin privileges always requires approval.
Connection strings are stored encrypted
They are stored encrypted with AES-256-GCM, and the plaintext never appears on screen, in logs, or in input to AI.
TLS is always verified over the internet
Direct connections have no code path that skips certificate verification. Private databases are reached through Cloudflare Tunnel.
Everything is on record
Queries run, judgments and their reasons, approvals, and execution results are all kept so you can trace them later.
What Scandock does
From design to operation
ScandockAI monitoring
AI decides what to watch, and keeps it running
AI learns how your service is structured from its schema and real data, then designs which values to watch, how often to fetch them, and how often to judge them. No more listing metrics or maintaining thresholds: AI keeps revising its monitors as it runs them.
- One monitor tracks several values together and judges them by how they relate
- AI retires monitors that have served their purpose, so no noise sources are left behind
DB
Payment outcomes
Watches successful and failed payments together to catch failures that cluster.
- Fetch interval
- 1 min
- Judgment interval
- 15 min
- Last fetched
- 14:03
- Last judged
- 13:48
SELECT count(*) FILTER (WHERE status = 'failed') AS failed, count(*) FILTER (WHERE status = 'succeeded') AS succeeded FROM payments WHERE created_at >= now() - interval '5 minutes'
Reasons and root cause
ScandockAI investigation
Every judgment comes with its reasoning
Instead of fixed thresholds, judgments draw on trends, past alerts and how they were resolved, and the facts built up in your project. When something looks wrong, AI follows the related data to where the cause lies, keeping a timeline of every query it ran and what came back.
- Runs judged as normal are kept with their reasons too, so you know what silence means
- Anything that catches your eye, you can ask to investigate further right from the same screen
Checking the spike in the payment error rate. First, let's see where the failures cluster.
Judged as an incident. Failures are concentrated on one card issuer, while order volume itself remains in its normal range.
Actions behind approval
ScandockActions
Actions run only after approval
Only actions you have registered in advance can be run. AI picks the action that fits the situation and proposes it, then runs it once approved. Anything that uses admin privileges always requires approval, and its rollback steps are recorded alongside it.
- Sending a notification is an action too, with its own message for each destination
- Proposal, approval, execution, and result can be followed as a single record
Temporarily extend the payment retry interval
Runs the registered action "Change retry interval". Rollback uses the same steps to restore the original value.
Works with your tools
ScandockIntegrations
Incidents and actions, inside the tools you already use
An integration is both where notifications go and an action AI can choose and run. Send escalations to chat, records to kintone, and events to your own systems by webhook, each where it belongs.
- Write a message for each destination, with placeholders that fill in the details
- Webhooks can be sent signed with HMAC-SHA256
Integrations
Slack
Notify a Slack channel
Teams
Notify a Teams channel
Google Chat
Notify a Google Chat space
Discord
Notify a Discord channel
LINE WORKS
Notify a LINE WORKS chat room
LINE
Send a message from a LINE Official Account
Chatwork
Post to a Chatwork room
kintone
Add a record to a kintone app
Send an email to a specified address
Webhook
Send JSON to any URL
Leave database monitoring to Scandock
Even if you're just gathering information, feel free to reach out.
For the databases behind your services
Whatever your industry, Scandock watches the databases your production services run on.
Watches that orders and payments keep flowing.
Examples of what AI watches
- Sudden changes in orders
- Payment failure rate and skew
- Backlog in stock allocation
Notices changes in per-tenant usage and delays in background work.
Examples of what AI watches
- Batch and job backlog
- Errors per tenant
- Sudden drops in usage
Checks that bookings and sign-ups proceed as usual.
Examples of what AI watches
- Bookings received
- Drop-off before sign-up completes
- Failed notifications
Core systems and internal tools can be watched the same way.
Examples of what AI watches
- Delays in data syncs
- Tables that stopped updating
- Long lock waits
Start monitoring in 4 steps
Nothing to install on your database servers. For databases inside your network, you run one cloudflared in the same network.
- STEP 1
Contact us
Contact us to discuss adoption. To try it right away, create an account and choose a plan.
- STEP 2
Register a database connection
Register a connection string. If the user can write, a read-only monitoring user is created on the spot. For Cloudflare D1 and BigQuery, just connect your account and pick one.
- STEP 3
AI builds the monitoring
AI reads the service's structure from its schema and real data, decides what to watch and how often, and adds it as a monitor.
- STEP 4
Receive alerts and operate
Alerts with the cause and a proposed fix arrive in Slack, email, LINE WORKS and more. Approve, and the registered action runs as is.
Plans and pricing
Choose by how many databases you watch and how finely they are judged. Judgments run per monitoring item, and investigations start only when something looks wrong, so normal operation costs no more.
Starter
Watch one production database as a team
¥298,000 billed annually (2 months free)
- Database connections
- 3
- Monitors
- 10
- Shortest judgment interval
- 30 minutes
- Automatic investigations
- 100 / month
- Members
- 5
- History retention
- 30 days
- Audit logs
- 30 days
Business
Judge databases across multiple services every 15 minutes
¥798,000 billed annually (2 months free)
- Database connections
- 10
- Monitors
- 50
- Shortest judgment interval
- 15 minutes
- Automatic investigations
- 500 / month
- Members
- 20
- History retention
- 90 days
- Audit logs
- 1 year
Enterprise
Contracted to your scale, governed with SSO and long-term audit logs
- Database connections
- Set by contract
- Monitors
- Set by contract
- Shortest judgment interval
- 5 minutes
- Automatic investigations
- Set by contract
- Members
- Unlimited
- History retention
- 365 days
- Audit logs
- 3 years
All prices exclude tax. A card is required to start the trial, and you are not charged if you cancel during the trial. Usage beyond the plan limits is billed per use.
Frequently asked questions
QWhich databases are supported?
PostgreSQL, MySQL, MariaDB, SQL Server, TiDB, Cloudflare D1 and BigQuery. Also works with Neon, PlanetScale, Azure SQL Database, Amazon RDS and Cloud SQL.
QCan you monitor internal databases that aren't exposed to the internet?
Yes. You run cloudflared on a machine in your network, and Scandock connects through Cloudflare Tunnel. You don't need to expose the database or open inbound firewall ports.
QCan my customers' personal data reach the AI?
No. Text values in your tables, such as names and email addresses, are masked before anything is sent to the AI. Numbers, dates, and table and column names used in the investigation are sent as they are.
QCould the AI modify my production database?
Monitoring and investigation use a read-only user, and queries the AI writes are limited to a single read statement. Actions that write run only as steps registered in advance, and only after approval.
QWhere is data stored, and what does the AI receive?
Accounts, connection details and incidents are stored in the TiDB Cloud Tokyo region. Monitor item value history is kept in Cloudflare Durable Objects, whose storage region is not fixed. What is sent to the AI, how long data is kept and our subprocessors are described on the security page.
QWhat do I need to do to get started?
Register one connection string (for Cloudflare D1 and BigQuery, just connect your account and pick one). AI designs the monitoring items, so there are no thresholds to set and no dashboards to build.
QHow do I pay?
Payment is by credit card (Stripe). Annual subscriptions can also be paid by bank transfer against an invoice. Prices exclude tax. For the Enterprise plan, please contact us.
Leave the database to AI
and focus on your business
Tell us about the databases you want to monitor or how you'd like to roll it out. You can also see the actual screens and example alerts in the demo, no sign-up needed.