Repository navigation
Image Analysis
Use dashboard captures or uploaded images as evidence for an Intelligence Gateway assessment. Panel 1.1.0 requires secure companion 1.1.0 or later; install the companion first and restart Grafana. Version 1.0.1 does not include image support. The Current View Exporter plugin does not need to be installed.
| Mode | Query setup | Image setup and action |
|---|---|---|
| Data only | Use this panel's queries or Dashboard → Source panel. | Leave images unattached; select Analyze. |
| Images only | No query results or Dashboard source required. | Turn Include query data with images off; preview images, select Use images, then Analyze images. |
| Data and images | Connect the query results needed for precise values. | Leave Include query data with images on; preview images, select Use images, then Analyze images. |
Every mode requires AI provider → Secure AI data source. This setting selects the model connection and is separate from Queries → Dashboard, which shares measurements. Attaching images permits an images-only assessment even when When no data arrives is set to Block.
- Set the dashboard time range and filters, and let panels finish rendering.
- Select Images to open Image inputs.
- Choose Capture dashboard, Capture selected panels, or Add images for PNG/JPEG uploads. Intelligence Gateway panels are excluded from dashboard capture by default.
- Review every preview, remove unwanted images, and check warnings and label legibility. Panel labels use their headings, not the full text body.
- Select Use images. This closes the preview without sending anything.
- Select Analyze images to submit the prepared images and prompt through the configured secure AI data source.
Remove discards one attachment; Remove all images discards all staged images. Clear analysis cancels an active request and removes the assessment, errors, and pending images, leaving measurements unchanged. Successful image assessments release their attachments; capture or attach again for another image assessment.
Auto-analysis never sends images and pauses while attachments or the image dialog are active. Images remain in browser memory, not dashboard JSON or persistent browser storage. Clearing cannot retract content already sent to a provider.
Capture uses the current-view exporter's rendered capture approach; it does not create a public Grafana snapshot or a print document. It does not deliberately refresh the dashboard or rerun queries. Scrolling to previously unrendered panels can trigger Grafana's normal initial queries, so this is not a universal zero-query guarantee. Collapsed, inaccessible, failed, or cross-origin content may be missing. Review the actual previews and warnings before sending.
Capturing another panel provides its image, not its underlying DataFrames. Use Grafana's Dashboard data source separately when a combined assessment needs precise query values. Models can misread axes, units, legends, small text, and numbers inferred from pixels.
- Turn on Enable image analysis (
enableVision, off by default) in the companion. Add the approved vision-capable model to both the normal model allow-list and Allowed vision models (allowedVisionModels). An enabled configuration with an empty vision list is rejected. Discovery and allow-list membership do not make a text-only model vision-capable. - OpenAI, LM Studio, Open WebUI, custom OpenAI-compatible Chat Completions, and the optional Grafana LLM app support structured image requests through their adapters. Actual model and parameter compatibility must be checked. Use the provider guide and perform an approved real-provider test with known visual facts.
- The Grafana LLM option uses a dedicated Grafana service-account token in
secureJsonData.bearerToken, not the user's SSO session. Itsbase/largealiases are configured ingrafana-llm-app. App 1.2.0 requires Grafana 13 and has a two-minute upstream timeout. - Copilot Studio has a Direct Line preview adapter with server-side
secureJsonData.directLineSecret, isolated conversations, bounded reply polling, and multipart image uploads. It is not corporate SSO, delegated Entra authentication, or app-only Direct-to-Engine access. See Copilot Studio Notes for completion-contract and tenant prerequisites. Never disable agent authentication to make it work; mock tests are not live tenant validation. - Limits: 8 images; 4 MiB decoded file size per image; 12 MiB combined; 4,096 pixels per side; 16 million pixels per image and 32 million combined; 18 MiB image-request JSON with non-image content still limited to 1 MiB. PNG/JPEG only. Providers can impose lower limits.
- Uploads are rasterized to remove embedded metadata such as EXIF/GPS; JPEGs may be re-encoded. Sensitive content visible in the pixels is not automatically redacted.
- Use synthetic or approved non-sensitive images. Dashboard viewing rights do not automatically permit sending screenshots to an AI provider. Provider retention policies apply after submission.
For Chat Completions providers, the lower panel/admin output cap applies. Hidden reasoning can consume that cap before a visible answer appears. Increasing only the administrator ceiling does not raise a smaller panel request. For a suitable local model, an example test budget is 8,192 hard output tokens in both settings, a 500-token soft answer target, and 300-second panel/backend deadlines. These are examples, not guaranteed model capabilities. The soft target is approximate; the default-limit option is not unlimited.
An output-budget failure, a timeout, and an unsupported image/model response are different problems. Copilot's agent owns generation policy; its hard output tokens and temperature are not controlled through Direct Line. A successful text completion or model-list request does not demonstrate visual understanding.
If controls still look unchanged after an upgrade, restart Grafana and fully reload the browser page. Dashboard Refresh refreshes queries, not the plugin runtime. See development cache guidance before clearing browser data.
See the complete image-analysis guide for exact endpoint examples, service-account permissions, limits, and validation boundaries.
Use panel 1.1.2 or later for tall-dashboard capture on Grafana 12.4.0. Companion 1.1.0 or later remains compatible. Stalled scrolling now reports an error; review any partial-capture warnings caused by image limits. Updating the separate exporter does not update this capture implementation.
Active capture survives temporary panel unmounting during dashboard scrolling (including Grafana 12.2). Navigation or a provider/model change cancels the capture. An unclaimed result is discarded five seconds after capture finishes; active sessions have a 70-second outer cleanup limit. Images remain only in browser memory and still require explicit review and submission.
- Home
- Reviewer Walkthrough
- Panel Setup and Configuration
- Connecting Data from Other Panels
- Image Analysis
- Copilot Studio and Enterprise Authentication
- Configuration: AI Providers
- LM Studio Setup
- OpenAI Setup
- Prompt Engineering Guide
- Security Considerations
- Secure Backend and Secret Storage
- Installation
- Release notes: 1.1.2
- Release notes: 1.1.0
- Grafana Compatibility and Certification
- Development and Contributing
- Roadmap