Monitoring plan
Bayes is Monte Carlo’s AI teammate. It observes your agents and the data that powers them, finds what breaks, proposes and verifies fixes, and gets better at your environment over time.
Bayes builds a monitoring plan for a domain: it analyzes the domain's data and proposes what to monitor. Instead of configuring monitors table by table, you point Bayes at a domain and it ranks and deep-dives the most important assets, then proposes a plan of monitors for you to review and enable. Nothing is deployed until you approve it -- you are the checkpoint at every step.
How it works
The monitoring plan runs scoped to a single domain. When you create a plan, Bayes ranks the domain's assets by criticality, deep-dives the highest-priority ones, and proposes a plan of monitors -- each with a plain-language reason for proposing it. It focuses effort on the assets that matter most rather than monitoring everything.
Proposed monitors are never deployed automatically. You shape a draft of the plan with Bayes first, then review the proposed monitors and enable all of them, or a subset by database, schema, table or monitor.
Creating a plan
Open Domains, select your domain, and open the Monitoring plan tab. Bayes builds the plan in three steps.
- Context & audience. Optionally describe what matters: priorities, sensitive periods, systems being migrated, or things to ignore. Context measurably improves the plan Bayes builds. Under Who gets alerts?, pick the audiences the plan's monitors will notify, or click Decide later to choose them when you enable monitors. If your account requires an audience on every monitor, you must pick one. Then click Draft plan.
- Review draft. Within a few minutes, Bayes drafts a plan: which tables to watch, which get deep checks, and which AI agents to include. Refine the draft in the Plan with Monte Carlo AI chat, or edit it directly: remove tables or agents, switch a table between Basic and Deep, or restore a critical table the draft left out. When the draft looks right, click Generate full plan.
- Monitors. Bayes builds the full plan of proposed monitors. Generation runs unattended -- Bayes works from historical patterns in your data, with no configuration needed from you during this window. This can take a while, depending on how many tables get deep checks. Nothing is enabled until you enable it. When the plan is ready, review the proposed monitors and enable them.
After you enable monitors, let Bayes keep monitoring healthy: in the domain's About tab, you can override the account settings for Auto-triage, Auto-troubleshoot (for example, for high and medium priority alerts) and Auto-tune (tuning). Which of these you see depends on your account's AI features.
Reviewing and enabling the plan
The plan groups the proposed monitors By database or By schema. On credit-priced accounts, it shows an estimated daily credit cost for the monitors not yet enabled. You stay in control of what actually runs.
- Enable monitors one at a time or all at once. Enable a single proposed monitor, a table, or a whole group, or use Enable all to deploy the entire plan.
- Track each monitor's state. Come back to the plan any time to see the state of every proposed monitor -- staged (proposed, not yet deployed), enabled, paused, or deleted. The plan is a living view of what Bayes proposed and what you have acted on.
- Refresh the plan. A plan reflects the domain as it was when Bayes built it. A plan that still has monitors to enable is marked outdated once it's more than 7 days old, or right away if it wasn't built from a draft you reviewed. While it's outdated, Enable all turns off until you refresh it -- you can still enable monitors one at a time. Use Refresh plan to have Bayes draft a new plan from the domain's saved context and the plan's audiences -- for example after you have added assets, defined use cases, or changed the context. The current plan stays available until you review the new draft. To start over with new context instead, use Plan again.
- Export as YAML. From the plan's actions menu (...), select Export YAML to export the plan and manage the proposed monitors in Monitors as Code, so they live alongside the rest of your version-controlled configuration.
Monitors Bayes creates from the plan are owned by the Agentic Platform Agent service user ([email protected]), so you can tell plan-created monitors apart from ones authored by hand. Each of these monitors keeps a View monitoring plan link back to the plan it was built from.
Before building the plan: structure your domain
Bayes plans per domain, so domain hygiene is the single biggest lever on plan quality.
- Organize domains with clear logic. Give the domain meaningful boundaries -- medallion layers, application or product lines, or data-mesh ownership. Structure it the way your team already thinks about the data.
- Assign every related asset. Add all related tables, dashboards, and other assets to the domain. Coverage is only as complete as the assets Bayes can see.
- Define your use cases. Use cases are generated automatically by Monte Carlo from your tables' usage, queries, and lineage, once that data has been ingested for at least 48 hours. They describe the business purpose of a group of tables and drive which assets Bayes prioritizes.
- Enable data sampling. Field-level (validation) recommendations work best with data sampling enabled on the warehouse, and with the
monitors/data-sampling/accesspermission for the user who generates the plan. Per-table and per-domain sampling restrictions also apply. Without sampling, Bayes has less data to profile tables and check its proposed rules.
Access control
The plan runs on the permissions of the user who creates it, so when Bayes builds a monitoring plan it can only see and plan against assets that user can access.
- Viewing and generating a plan needs permission to create or draft monitors in the domain (
monitors/management/writeormonitors/management/propose). Users who can only view monitors can't see the plan. - Enabling monitors from the plan needs permission to edit monitors in the domain.
- Saving context to the domain needs
settings/domains/edit. Context is optional; users who can draft a plan but can't save context can still add context for that one plan. Context saved to the domain, or to the account in Settings > AI Agents (settings/edit), applies to every user's plans. - Bayes' triage and troubleshooting aren't required. Turning them on after you enable the plan is recommended, not a prerequisite.
Working on a plan with others
- A domain has one shared plan. Everyone who opens the domain's plan sees the newest completed plan.
- Drafts are personal. Each user works on their own draft and conversation with Bayes. A draft becomes the domain's plan when it's generated.
- The newest plan reflects whoever generated it. If a user with access to fewer assets, or without data sampling access, generates a plan, that narrower plan becomes the one everyone sees until someone with broader access generates a new one. Earlier plans that proposed at least one monitor stay available to switch back to.
If something looks off
| Symptom | What to check |
|---|---|
| Plan is taking a long time | Large domains take longer -- Bayes spends up to 1-2 minutes per qualifying table, so a domain with many tables can run for many hours. Also verify when asset ingestion started; Bayes may need several days of historical data to generate an optimal plan. |
| Plan seems sparse | Confirm all relevant tables and metadata are assigned to the domain -- Bayes only plans against assets it can see. Check that data sampling is enabled if you expect field-level monitors. |
| Questions on the plan | Reach out to your Monte Carlo account team and we'll review the plan with you. |
Security & data privacy
For more detail on security and data privacy, see the AI Features and Technical Information documentation.
Feedback
Have feedback or requests on the monitoring plan? Reach out via in-app support.
Details and edge cases
Generating a plan while another is in progress. Only one plan can be generated at a time per domain. While one is in progress, Monte Carlo won't start another for that domain.
Enabling the same plan twice. Only one enable can run on a plan at a time, so enabling twice doesn't create duplicate monitors.
Updated 1 day ago
