Resource Generation¶

Obsy AI Copilot can use natural language and real metadata from the current workspace to generate dashboard drafts and Pipeline parsing rules.

Create and Modify Dashboards¶

Copilot can create a dashboard from natural-language requirements and refine charts, groups, and presentation goals over multiple turns. The flow consists of requirement confirmation, draft generation, preview generation, and saving.

Before You Start¶

• The current account must have permission to create dashboards.
• The current workspace must contain data sources and fields that match the target scenario.
• Define the monitored object, data scope, groups, key metrics, and filter dimensions whenever possible.

If a request only says to create a dashboard without identifying what to monitor, Copilot asks for more scope before generating a draft. If the account lacks permission, the system asks you to contact an administrator and does not enter the creation flow.

Start the Creation Flow¶

Generate and Refine the Draft¶

After entering the flow, describe what the dashboard should monitor or analyze. A complex request can also specify:

• Metrics, logs, events, traces, or other data domains to use;
• Dashboard groups and the concern for each group;
• Chart count, preferred chart types, or required analytical views;
• Filter dimensions or default values such as host, service, and env;
• Charts or data scopes that should be excluded.

Copilot discovers data sources, fields, and tags that actually exist in the current workspace, then generates a dashboard draft. The draft includes the name, description, visibility, view variables, and chart cards. At this stage, final queries have not yet been generated and no dashboard has been saved.

Continue refining the draft through the conversation. For example, ask Copilot to move error distribution into the first group, remove a host-count card, or add a P95 latency trend grouped by service. Identify the target chart, group, or result instead of only asking it to "optimize the draft."

If available data sources, fields, or correlation tags cannot support a chart, Copilot reduces the unsupported content or identifies the data gap instead of filling the draft with unconfirmed fields.

Preview and Save¶

When the draft is ready, select Confirm Dashboard Generation to open the preview. The system generates the actual configuration for each chart and displays completed chart count, progress, and failures.

After generation, you can:

• Inspect chart data and adjust charts from the preview;
• Select Return to AI Conversation to refine the requirements and generate again;
• Select Save Dashboard after confirming that the content is correct.

The dashboard appears in the list only after it is saved. If some charts fail to generate, the page asks you to return to Copilot and adjust the draft; do not rely on failed or unchecked charts.

Recover a Draft¶

If you leave during the draft stage, the AI dashboard entry lets you continue the previous draft or clear it and start again.

A dashboard generated and saved through Copilot displays an AI indicator in the dashboard list.

Generate Pipeline Parsing Rules¶

Use this capability when logs or line-protocol data need to be split into structured fields and you do not want to write a Pipeline rule from a blank page. Copilot generates a rule from a representative sample and the desired target fields.

Common inputs include:

• JSON logs;
• key=value text;
• Plain log text with stable boundaries;
• Complete JSON Point or line-protocol samples containing Point metadata;
• Log messages that contain embedded JSON or another structured segment.

Generate and Validate a Rule¶

1. Provide a representative production-format sample in a Pipeline flow that supports AI generation.
2. Describe the fields to extract, target field names, data types, and which fields should be tags.
3. State explicitly when all fields should be extracted, when the source timestamp should become the data timestamp, or when outer Point fields should be retained.
4. Validate the generated rule against the sample, inspect the extracted values, and then save and apply it through the regular Pipeline flow.

When the sample contains changing values, preserve the real structure instead of replacing boundaries with fixed IDs, hosts, or timestamps. If no target field names are provided, Copilot uses stable English snake_case names.

Rule Output and Activation¶

Copilot returns a Pipeline parsing rule and selects JSON, Grok, key-value splitting, or Point field access according to the input structure. The generated rule does not remove source data automatically, change the measurement, or turn fields into tags unless the request explicitly requires it.

The rule affects later processing only after it has been validated, saved, and enabled through the existing Pipeline mechanism. If the sample does not represent the actual data format, some fields may not be extracted; provide a better sample and validate again.