Skip to main content

Connections

Version: 8.19 | Last updated: 2026-11-10

Bluebox connects to external services to extend what it can investigate and where it can assist. This page covers the AWS DevOps Agent, Azure SRE Agent, Kiro, and GitHub Copilot integrations.

Manage connection access​

Every connection has an owner: the workspace member who created it. A new connection is private to its owner by default. Other workspace members can see that the connection exists and view its state, but they cannot reconnect it, re-authorize it, change its configuration, or delete it.

The connection's owner can fully manage it: reconnect or re-authorize it, change its settings, change who it is shared with, and delete it. A workspace owner, or any member the owner has granted edit access, can also reconnect, re-authorize, change, and delete the connection. Changing who a connection is shared with is reserved for the connection's owner and the workspace owner; a member with edit access cannot change its sharing.

If you want another member to help manage a connection you own, share edit access with them explicitly. Sharing a connection without edit access lets other members view it but not change it. If the owner also turns off read access, other members still see the connection's name and state in the list, but not its settings, and they cannot open its configuration.

This applies to every integration on this page.

Use the search box above the integrations table to filter by name, type, or MCP tool name.

Connection audit log​

Bluebox records an entry in the connection audit log when it invokes a tool through a connection. It also records an entry when someone changes a connection. The log shows who used or changed a connection, when, and with what result.

These changes create an entry:

  • Connection created
  • Authentication changed
  • Connection disabled
  • Connection enabled
  • Connection deleted
  • Sharing updated
  • Tool policy updated

Each entry shows:

  • Time — when the call or change happened
  • Used by — the display name or email of the workspace member who triggered the call or change. If the call was made by an automated routine with no user context, this column shows Automation.
  • Connection — the name of the connection that was used. If the connection was deleted since, this column shows the connection ID.
  • Event — a tool invocation or the type of connection change
  • Tool — the name of the tool that was called. This column shows — for connection changes.
  • Outcome — whether the call or change succeeded or failed
  • Error code — the failure reason, when the call or change failed

Entries for a deleted connection stay in the workspace-wide audit log. The log keeps up to 10,000 entries. When it is full, Bluebox removes the oldest tool invocation entries before it removes connection changes.

Who can see the audit log​

The per-connection audit log is available to:

  • The connection's owner
  • Any workspace member who has read access to the connection
  • Workspace owners

Open Settings → Integrations, expand a connection, and scroll to the Audit log section.

The workspace-wide audit log aggregates usage across all connections. It is visible only to workspace owners. It appears at the bottom of the Settings → Integrations page, below the connections table.

Clear the log​

Workspace owners can clear the entire audit log. Select Clear log in the audit log panel. This action cannot be undone.

AWS DevOps Agent​

The AWS DevOps Agent integration lets Bluebox run investigations against your AWS account.

Once a connection exists, only its owner or a workspace owner can change or remove it. Other workspace members can view its connection state but cannot modify it (see Managing connection access above).

When an investigation runs, Bluebox evaluates registered AgentSpaces and routes sub-investigations to relevant ones. Each AgentSpace contributes root-cause analysis and mitigation proposals incorporated into the final result.

Setup​

Open Setup → Integrations → Add integration and select AWS DevOps Agent. The wizard has three steps:

  1. Configure—Enter a display name and select an AWS region.

  2. Setup IAM—Download the setup script (bash, Terraform, or CloudFormation) and run it with AWS admin and iam:* permissions. The script registers Bluebox as an OIDC provider in your account, creates an IAM role with a scoped trust policy (pinned to the integration's issuer, audience, and subject), and attaches the following permissions:

    Permission groupActions
    Discoveraidevops:ListAgentSpaces
    Task managementaidevops:CreateBacklogTask, aidevops:UpdateBacklogTask, aidevops:GetBacklogTask, aidevops:ListExecutions
    Investigationaidevops:ListJournalRecords, aidevops:SendMessage, aidevops:CreateChat

    The script prints the role ARN when it completes.

  3. Verify—Paste the role ARN. Bluebox stores the connection and validates the ARN format. On success, the integration becomes active. If the ARN format is invalid, the wizard shows a specific error so you can correct it and retry without starting over.

Resume a pending connection​

If you close the wizard before completing step 3, Bluebox attempts to remove the pending connection automatically. If the removal fails (for example, due to a network error), the connection stays in a pending state. The integration row shows pending in the status column. You cannot register AgentSpaces until the connection is active.

To resume setup, select Complete setup from the row's overflow menu. The wizard reopens at step 1 with the name and region pre-filled. Run the setup script and paste the role ARN to activate the connection. If you cancel again, the pending connection is removed automatically.

AgentSpaces​

Once the integration is active, select Discover & register to query ListAgentSpaces and see which spaces are available in your account. Select the spaces you want to register with Bluebox.

Fill in the description field for every space you register. Bluebox uses these descriptions to decide whether to involve a given AgentSpace in an investigation. A space with no description is never selected. On a space with no description yet, use Ask AgentSpace to query it directly: the space describes its own purpose and Bluebox drafts the description for you. On a space that already has a description, use Regenerate with AI to refresh it.

Manage an integration​

  • Re-display the setup script: Open Configure → Show setup script. The per-integration client ID is stable for the lifetime of the connection.
  • Edit a space description: Select the registered space, update the description field, and save. Changes take effect on the next investigation.

Remove an integration​

If the connection is active, select Disable on the integration row first. Once the connection is disabled, select Delete. If any investigations are still running, Bluebox rejects the delete. Wait for them to finish, then retry. Bluebox removes the connection and all registered AgentSpaces. The OIDC provider and role in your AWS account are not removed, so remove them manually once nothing else references them:

aws iam delete-role --role-name <role-name>
aws iam delete-open-id-connect-provider --open-id-connect-provider-arn <arn>

Azure SRE Agent​

The Azure SRE Agent integration lets Bluebox run investigations against your Azure subscription.

Once a connection exists, only its owner or a workspace owner can change or remove it. Other workspace members can view its connection state but cannot modify it (see Manage connection access above).

When an investigation runs, Bluebox evaluates registered SRE Agents and routes sub-investigations to relevant ones. Each SRE Agent contributes root-cause analysis and mitigation proposals incorporated into the final result.

Setup​

You need the Azure CLI installed and signed in with permissions to create app registrations and assign roles on the target subscription.

Open Setup → Integrations → Add integration and select Azure SRE Agent. The wizard has three steps:

  1. Configure—Enter a display name and select an Azure region.

  2. Setup Entra—Download the setup script and run it. The script creates an Entra app registration with a federated identity credential (no client secret is stored), creates a service principal, and grants the following roles on your subscription:

    RolePurpose
    ReaderDiscover SRE Agent resources in the subscription
    SRE Agent Standard UserCreate and monitor investigation threads

    The script prints your Tenant ID, Client ID, and Subscription ID when it completes.

  3. Verify—Paste the complete script output. Bluebox extracts the three identifiers automatically. On success, the integration becomes active. If extraction fails (typically because the output is incomplete), the wizard shows a validation hint so you can re-paste without starting over.

SRE Agents​

Once the integration is active, select Discover & register to list SRE Agent resources in your Azure subscription. Select the agents you want to register with Bluebox.

Fill in the description field for every agent you register. Bluebox uses these descriptions to decide whether to involve a given SRE Agent in an investigation. An agent with no description is never selected.

Manage an integration​

  • Re-display the setup script: Open Configure → Show setup script. The per-integration client ID is stable for the lifetime of the connection.
  • Edit an agent description: Select the registered agent, update the description field, and save. Changes take effect on the next investigation.

Remove an integration​

If the connection is active, select Disable on the integration row first. Once the connection is disabled, select Delete. If any investigations are still running, Bluebox rejects the delete. Wait for them to finish, then retry. Bluebox removes the connection and all registered SRE Agents. The Entra app registration is not removed, so delete it manually once nothing else references it:

az role assignment list --assignee <client-id> --query "[].id" -o tsv \
| xargs -r -I{} az role assignment delete --ids {}
az ad app delete --id <client-id>

Remove the role assignments before deleting the app registration. Azure does not clean up RBAC assignments automatically, so they become orphaned and remain visible in the subscription until explicitly deleted.

Cloud monitoring​

Selective preview

A cloud monitoring connection gives Bluebox read-only access to an AWS account, an Azure tenant, or a Google Cloud project so it can list what you run there, estimate what monitoring it would cost, and produce the monitoring configuration for it. It is separate from the agent integrations above: nothing in it runs investigations, and nothing in it can change anything in your cloud.

Once a connection exists, only its owner or a workspace owner can change or remove it (see Manage connection access above).

Connect an AWS account​

You need a shell with credentials for the account you want monitored and permission to create IAM roles. AWS CloudShell has both.

Open Setup → Integrations → Add integration and select AWS monitoring. The wizard has three steps:

  1. Configure — Enter a display name. Nothing else is asked for: the region is whichever one you run the script in, and the script reports it back.

  2. Run in AWS — Download the script and run it. It creates two read-only roles: one Bluebox assumes to read your account, and one the observability backend assumes to collect monitoring data. Neither can create, change, or delete anything. Running it again is safe — it converges rather than creating a second set.

  3. Verify — Paste the complete script output. Bluebox reads the two role ARNs and the region from it, and says so if any is missing, so you can re-paste without starting over. It then reads from your account and reports what it could see — how many accounts, by name. That check is what tells you the connection works; saving it only records what you pasted.

Two things the script cannot do for you, both of which it prints at the end:

  • Dependency arrows in the architecture view come from AWS Config's recorded relationships. Without a configuration recorder, the view still groups your applications but shows no dependencies between them.
  • Direct cost attribution needs a cost allocation tag activated in Billing → Cost allocation tags. Until it is, that figure is reported as unavailable rather than estimated.

Connect an Azure tenant​

You need the Azure CLI signed in with permission to create app registrations and assign roles. Azure Cloud Shell has both.

Open Setup → Integrations → Add integration and select Azure monitoring. The steps match the AWS flow; the script creates one app registration with federated credentials, so no secret is stored, and prints a tenant ID and a client ID to paste back.

The app registration it creates carries a suffix derived from the connection, so two connections never share one and the script never adopts an app someone else created under the same name.

Connect a Google Cloud project​

You need the Google Cloud CLI signed in with permission to create a service account, grant roles, and create a workload identity pool. Cloud Shell has both.

Open Setup → Integrations → Add integration and select Google Cloud Monitoring. The steps match the AWS flow. The script creates a service account and a workload identity pool with two providers, one for collection and one for Bluebox. It does not create a service-account key. Paste the service account, both provider names, and the project id it prints.

Measured cost needs a BigQuery billing export. The script cannot turn that on. Enable it under Cloud Billing → Billing export. The export is not retroactive, so a table created today cannot price monitoring that already ran.

Finish setup​

When you select Done in the wizard, Configure opens on the Setup tab and Bluebox takes it from there:

  1. Wait for access — Bluebox checks the account, less often as time passes, until the cloud role takes effect, and shows how long it has waited. After 10 minutes it stops, says what to check, and offers Try again.
  2. Discover resources — Bluebox lists what you run.
  3. Estimate the cost — Bluebox prices the monitoring for the suggested preset.

None of these steps writes anything to your environment or your cloud. Discovery continues if you close the window. The access check runs while Configure is open, and Setup picks up where it was when you open it again.

Then Review and apply asks for three decisions:

  • Monitoring preset — Troubleshooting is suggested. The estimated cost follows your choice.
  • Monthly budget limit — pre-filled with the estimate plus 20% headroom, rounded up to the next $10. The agent applies changes only while the estimate stays within it. You can still apply over it yourself; Bluebox warns you first.
  • Agent may apply changes — Ask me first (the default) or On its own, within budget.

Select Apply. Bluebox saves the three choices, installs the monitoring extension in your environment if it has none, and writes the configuration. Only the connection owner or a workspace owner can apply.

Choose what the connection may write​

Both wizards show Enable autonomous log forwarding, off and unavailable, because Bluebox does not forward logs yet. The scripts grant read access only: no event hubs, diagnostic settings, or tags are created or changed.

You can see what a connection granted at any time: open Configure → Show setup script. Bluebox rebuilds the script from what the connection actually records, so it always reflects the current grant rather than a saved copy.

Understand a partly connected state​

A connection has two independent halves: the access Bluebox uses to read your cloud, and the access the observability backend uses to collect monitoring data. The first is enough for discovery, cost estimates, and the architecture view, and it becomes usable as soon as the wizard finishes. Bluebox reports each half separately rather than a single status, so a connection that is working for some things and not others says which.

Both halves are set up by the one script. The collecting half is anchored to an identifier Bluebox creates before it renders the script, which is why the display name is the only thing the first step asks for — everything else has to exist before there is a script to run.

Azure role assignments are also granted on a best-effort basis: the script counts any that did not go through and reports the total. The wizard passes that count on, and you can continue — monitoring is partial until you fix the causes the script listed and run it again.

Re-running the Azure script is safe for everything except the app registration: role assignments, credentials, and the resource group are only created when they are missing. The app registration is deliberately different. The script will not reuse an application it did not create in that run, even one that looks like its own, because nothing it can check proves who registered it — so if the identifier is already taken it stops and names the application instead of granting it anything. If your own earlier run created it, clear its role assignments in every subscription first and delete the application last, then run the script again — the message includes the commands. Both parts of that order matter. Azure does not remove an application's role assignments along with it, and once the application is gone the assignments can no longer be found by its id, so deleting it first leaves entries behind that are awkward to track down. And the script grants on every subscription your login can see, and on a resource group inside each, so a cleanup covering only the current subscription — or only its top level — leaves the rest; the commands loop over every subscription and cover every scope beneath it. If you did not create the application, do not grant anything and report it.

Set a budget, a preset, and tag filters​

Open Configure on a cloud monitoring connection to change four things. None of them is asked for while you are connecting, because none of them can be answered sensibly before you see what is in the account. They become available once the connection finishes setup — before that the connection still holds the state the setup script needs, and a setting saved into it would be discarded when setup completes.

Agent may apply changes is the first control and decides whether the agent asks you before it acts, and nothing else. Choose Ask me first or On its own, within budget. It grants no access either way. Setup saves Ask me first unless you choose otherwise. With Ask me first, the daily run is paused, and Bluebox refuses the agent three things on this connection: applying the configuration, changing these settings, and installing a missing monitoring extension while it previews the configuration. It tells you to apply from Configure, or to switch the connection to On its own. Everything it can only read stays open to it. With On its own, within budget, the agent applies the configuration only while the estimate stays within the monthly budget limit, and applies nothing when no limit is set. You can still apply over the limit yourself from Configure after a warning.

Monitoring preset decides which services are monitored and which of their logs are kept. Bluebox does not forward logs yet, so today a preset decides the services.

  • Troubleshooting — every supported service discovery found.
  • Auditing — a minimal set for audits: firewalls and web application firewalls, gateways and load balancers, certificates and key vaults, backup, storage, and databases.
  • All-in — every supported service discovery found and all of its metrics, including the custom metrics found for service types that have no built-in set.

All-in is the default for a connection that never chose a preset. A preset that matches none of the services discovery found is refused with a message. Each preset card shows what it would cost per month.

Below the presets, Settings lists every service and every region discovery found, each with a checkbox, its resource count and its monthly cost. Clear a checkbox to stop monitoring that service or region, or tick it to add one the preset leaves out. Your choice wins over the preset and is kept until you change it. Choosing another preset after that asks first, because it replaces your service selection.

Monthly budget limit is what this connection may cost per month. The agent applies changes only while the estimate stays within it. Leaving it empty means no budget, and emptying a field that had a figure in it removes that budget. Entering zero is a different statement — it means spend nothing — so an empty field is never read as a budget of zero.

Tag scope narrows which resources are monitored, which is the main lever for making ingest cheaper without dropping whole services. Include keeps only resources with any one of the listed tags. Exclude drops resources with any of its tags, and an exclusion wins over every tag you include. Both AWS and Azure accept exclusions. Once discovery has run, Settings lists the tags it found; select + next to a tag to include it, or − to exclude it, and remove a tag from either list with its delete button. Empty means monitor everything. The cost estimate does not narrow its figures by tag, so a filtered scope costs less than it shows.

Each change shows as a pending change, with what it adds to or saves on the monthly estimate and whether the new figure stays within the budget limit. Discard changes drops them. After the first apply, Save and apply saves what you changed and applies it in one step.

When someone turned a service on or off directly in your observability backend, Settings names it and asks whether to keep that change or turn it back. Save and apply stays unavailable until you decide, so the next apply does not silently undo the change. Changing any of these is restricted to the connection's owner or a workspace owner, because they decide what the workspace pays to ingest and when the agent stops. Other members can see the values.

Ask the agent to work with a cloud monitoring connection​

In a chat or an investigation, the agent can see the cloud monitoring connections you can read — your own, and those another member shared with you for reading — and can verify the connection, run discovery, map the architecture, estimate the cost, preview the configuration, and read the settings. A connection another member keeps private is not visible to it. In auto mode it can also change the settings and apply the configuration, and it is told to preview first and to say what the apply will change and cost. Discovery, mapping and apply run in the background; the agent follows them and reports when they finish. It sees a summary of the latest discovery, never the full report.

Run discovery and see what it found​

Open Configure on a cloud monitoring connection and select Run discovery. Discovery reads what you run in the account or tenant and writes nothing.

A full sweep takes several minutes and keeps running if you close the panel. While it runs, the panel says so and shows how many regions (AWS) or subscriptions (Azure) it has scanned out of the total; when it finishes, it lists what it found — subscriptions or accounts, service types, custom-metric specs, tag keys, and regions.

A sweep can finish and still be incomplete. When something could not be read — a Resource Graph table your tenant rejects, or a subscription the credential cannot see — the result is marked Succeeded with warnings and each warning is listed. Read that as a partial inventory: the cost estimate and the configuration below cover only what discovery could see. Running discovery again after you fix the cause is safe.

If a sweep fails, the panel names why. Start another one once you have addressed it.

See how your resources fit together​

The Architecture memory section of Configure maps the applications in the account or tenant: which resources belong together, the lane each one sits in (front end, application, compute, data, storage, networking), how their virtual networks, subnets, and security groups are laid out, and which resources depend on each other. Select Map architecture to start it. The section is titled Architecture memory, and its subtitle states how many of your applications it draws. It draws at most the five largest applications by resource count and summarizes the rest below. Mapping runs as its own sweep, separate from discovery, takes several minutes, and writes nothing.

Each application says why its resources were grouped. Bluebox looks first for an application tag (such as app, application, service, or project), then for strong connections between resources, and for the rest it falls back to the resource group, ignoring shared infrastructure groups such as hub, network, and monitoring. Security groups state in words who can reach their ports, such as Open to the internet or Internal only. Dependencies show as a graph or a list: a solid line is a relationship read from the resource itself, a dashed line one Bluebox inferred. Resources counted rather than drawn, and infrastructure no application claims, are totaled at the bottom. Each application's resources and networking are collapsed under Resources and networking, so the dependency graph comes first. Select Map architecture again to refresh the map after your resources change.

Every detail view ends its header with a Last run line that says when its action last ran; hover it for the exact time.

See what ran​

The Activity tab lists this connection's discovery, architecture, apply, and actual-spend runs, newest first, with each run's status, when it started and finished, and the error when it failed or is being retried. Run discovery again starts a new discovery. After the first apply, Show the configuration diff against your observability backend lists each field the next apply would change. Below it, the audit log records the tool calls made through the connection.

Review the cost estimate​

The estimate is a projection from the inventory discovery stored, not a reading of your bill, and no billing data is read. It splits the monthly figure into what the cloud provider invoices and what your monitoring consumption costs, one line per cost driver.

Each line states how far it can be relied on:

  • Exact — derived from the inventory.
  • Modelled — calculated from a model, because the real volume is unknown until monitoring is on. Log volume is the usual case.
  • Lower bound — the least this can cost, not the most.

Under the table, Bluebox says whether the total fits the monthly budget you set and by how much it misses if it does not, and names the preset, the tag scope, and the service count the figure was computed from. It also gives the age of the inventory it used, because a figure from a stale sweep is a figure about a cloud you no longer run, and the time the estimate was computed. Re-run recomputes it with your current settings without rediscovering; run discovery to refresh the inventory itself. Set as monthly budget saves the estimate plus 20% headroom, rounded up to the next $10, as this connection's budget; it is available to the owner or a workspace owner.

See what your cloud has billed this month​

Next to the projection, the cost view has a second block, Actual spend this month. It is a different thing from the estimate: it is read from your cloud's billing data, for the period from the first of the month (UTC) until now, and it is never computed from the inventory. The two blocks are never combined, and Bluebox does not compare a partial month with the monthly estimate.

Select Read actual spend to start it, and Refresh to read it again. Reading runs in the background, and the block shows it running, then the result, or the reason it failed. Only the connection owner or a workspace owner can read it, because it is the owner’s cloud billing data; other members see why the button is unavailable. Reading it changes nothing in your cloud and nothing about what you pay.

Each line is one cost driver, and each says in words how far its figure reaches:

  • Billed on this connection’s own resources — the cost of infrastructure Bluebox set up for this connection, found through the monitoring configuration Bluebox applied. Until you apply a configuration, Bluebox cannot tell which resources are the connection’s, so those lines are listed under Not counted with what they cost, instead of being guessed.
  • Whole subscription or Whole account — a meter the whole cloud account shares, such as metric reads. The figure includes spend that is not this integration’s, because there is no earlier period to subtract. Read it as an upper bound.

Under the lines, the total is what was billed on them this month so far. Monitoring-related spend that fits no line is shown separately and is not in the total. Cloud billing data arrives up to 24 hours late, so the block is marked Partial and today’s spend is not in it yet. A line saying that consumption is not available appears because this version does not read what your monitoring consumes in your observability backend; the block shows the cloud invoice only, and does not fill the gap with an estimate.

If the read fails, the reason is stated:

  • Azure — Cost Management refused the connection’s identity. Grant it Reader or Cost Management Reader on the subscription. The setup script assigns Reader on every subscription it can see. On an Enterprise Agreement, an enterprise admin must also enable charge visibility for subscription users; no setup script can do that for you.
  • AWS — Cost Explorer refused the connection’s role. Enable Cost Explorer for the account; the role the script creates already allows the billing read.

A read that fails for any other reason is retried a few times, then reported as failed. Runs, including failed ones, are listed in the Activity tab as Actual spend.

Review the monitoring configuration before applying it​

Select Preview configuration to see the monitoring configuration your settings and the latest discovery imply. The preview compares that document with the configuration your observability backend already holds for this connection and lists each field that would change, with the members added and removed named individually: added members are green with a plus, removed members red with a minus. Before the first apply there is none, so every member the document enables is listed. The preview changes no configuration, and there is nothing to paste. If your observability backend does not have the cloud's monitoring extension yet, the preview first installs the latest version from the Hub and says which version it installed. Expand Document that would be applied to read it in full.

On AWS, the configuration monitors only the regions where discovery found a resource that your observability backend can monitor, a CloudWatch namespace you chose to collect as custom metrics, or a CloudWatch log group. A region that holds only AWS defaults, such as the default VPC, or resources it cannot monitor, such as CloudFormation stacks, is left out. If no region qualifies, the preview and the apply refuse with No monitorable resources found and write nothing, and they do not install the monitoring extension. When an existing configuration lists more regions, the next apply removes the ones that no longer qualify. A region that discovery could not fully read stays in the configuration until a later discovery reads it. A discovery from before region selection cannot narrow the list, so the preview and the apply ask you to run a discovery again.

Apply configuration writes it into your observability backend. Only the connection's owner or a workspace owner can preview or apply, because the preview can install the monitoring extension and the apply changes what the workspace pays to ingest and where its monitoring points. Other members see the estimate, and see the preview and apply controls disabled with that reason next to them.

Applying runs in the background, and the panel shows it running. Bluebox finds the configuration it already wrote for this connection, so when the apply finishes the panel shows what it actually changed in your environment: whether it created the configuration or updated it, and which feature sets, regions, and tag-enrichment keys it enabled and which it disabled. On the first apply there is no earlier configuration, so everything in the document is new. Applying again updates the same configuration rather than creating a second one. Bluebox reads the configuration back after writing it, and reports the apply as failed if what your observability backend holds does not match what was written.

If the extension is still missing when you apply, the apply installs it before it writes the configuration, and the result says which version it installed. The preview and the apply need a credential allowed to read and write extension configurations and to install extensions. If one of those permissions is missing, Bluebox names it instead of writing anything.

Check that data arrives​

After an apply succeeds, the Overview tab of Configure leads with one status and one line: how many service types and regions are monitored, and when data last arrived. Chat about it and Investigate open a chat about the connection. Below that:

  • The actual cloud bill this month, the estimated monthly cost, and the budget limit.
  • Needs attention, only when a signal is Degraded, Stale, or No data, or the estimate is over the budget limit. Each item has its own Chat about this and Investigate.
  • Data, with the monitored services. A green dot means metrics arrive. The line below names the signals that are not set up.
  • Data details, closed by default, with every signal row.
  • Rollout, with the six steps. View opens a step's result and Run again runs it. The header says whether the daily run is on, and holds the Daily run switch and Run all now.

The status is the worst state of the signal rows: Receiving, Degraded, Stale, or No data. A row that is Not set up or Not confirmed does not change it. When Bluebox cannot read your observability backend, the status is Not confirmed, and Try again reads it again. Below the status are the number of accounts or subscriptions, regions, resources, and service types, and when data last arrived. Everything on the tab counts only the accounts or subscriptions of this connection. A connection with more than 1,000 resources shows Incomplete read instead of counts. An account or subscription shows its name when your observability backend has one, and its ID otherwise.

Each row is one signal:

  • Connection and monitoring configuration applied: whether the configuration is in place and the connection is healthy.
  • Metrics: whether the running resources of this connection sent metrics in the last 20 minutes. It shows Degraded when some of them sent none, and Not confirmed when none of its resource types has a metric to check.
  • Pollers · 24 h: what the collecting pollers reported for this connection's configuration in the last 24 hours. Errors make it Degraded, and the last error is shown under the row.
  • Discovery: when discovery last ran. After two days it is Stale, and Run discovery starts it again.
  • Cloud alerts, the provider's health events, Activity log or CloudTrail, and Logs show Not set up. Bluebox does not read them for this view yet.

Select Check now to read the rows again. Every member who can read the connection sees the same rows.

Run the rollout every day​

On the Overview tab of Configure, the Daily run switch in the Rollout header adds a routine that runs the rollout for this connection once a day. It is a separate step you take. Bluebox never adds it for you.

Each run repeats discovery, estimates the cost, previews the configuration, and applies it when the checks hold. It ends with a one-line result, for example APPLIED or NOT APPLIED (no budget). The routine never changes the preset, the onboarding mode, the monthly budget, or the tag filter. Those stay as you set them. It may switch off monitoring that your environment no longer needs. The names it switched off are in the result of the run.

To add it, turn on Daily run. Turn it off to remove the routine. Run all now starts a run at once; without a routine it starts discovery. The switch is unavailable, with the reason in its tooltip, in these cases:

  • You are not the connection owner or a workspace owner.
  • No monthly budget is set. Set one in Settings first.
  • Routines is not available in your workspace.

If the connection is set to Ask me first, you can still add the routine. It starts paused, and the Rollout header says so. A paused routine starts no runs, so it uses none of your workspace's run allowance.

You own the routine, and it starts as Only me. To let everyone in your workspace open it, read-only, change its sharing on the Routines page. Members who can read the connection see its status in the Rollout header in either case. Only you, or a workspace owner, can edit, pause, or remove it.

The Rollout header shows whether the daily run is on or paused. Switching the connection to Ask me first pauses the routine, and you resume it yourself on the Routines page after you switch back to On its own. A run that fails does not pause the routine.

Select Remove to delete the routine. If someone deletes it on the Routines page instead, the card shows that there is no routine, and you can add one again.

Remove a cloud monitoring connection​

If the connection is active, click Disable on its row, then Delete. Bluebox removes the connection, the monitoring configuration it applied, and the collecting-side record it created, but not what the script created in your cloud — clean that up yourself once nothing else references it: on AWS, delete the two roles; on Azure, delete the role assignments and then the app registration, in that order.

Kiro​

Bluebox supports both the Kiro IDE and Kiro Web.

Kiro IDE​

Running bluebox setup detects the Kiro IDE and installs Bluebox skills automatically.

Skills are loaded on demand. Kiro does not load them automatically at session start. To give Kiro Bluebox context in every session without having to ask, add a steering file:

bluebox skills get bluebox-overview > .kiro/steering/bluebox-overview.md

Commit this file to your repository so all team members get the same context automatically.

Kiro Web​

To add production-insight capabilities to Kiro Web, complete the following four steps. Step 1 requires an AWS account administrator; you can do the rest yourself.

1. Enable the required Kiro permissions

Running the Bluebox CLI from Kiro Web depends on Kiro settings that only an AWS account administrator can change. In the AWS Management Console, open Kiro → Settings → Kiro and enable:

  • Autonomous agents—lets Kiro run the Bluebox CLI (downloading it and running bluebox ask) without asking you to approve each command.
  • Web search and web fetch tools—lets Kiro download the Bluebox CLI from the links below.

Without these, Kiro Web cannot download or run the Bluebox CLI, and the bluebox-production-query steering has no effect.

2. Add steerings

In Kiro Web → Settings → Agent → Steering, create two steerings:

Steering 1—Download Bluebox CLI

Depending on your OS and architecture, download the Bluebox CLI from:
- Linux (x86-64): https://app.bluebox.ai/download/bluebox?os=linux&arch=amd64
- Linux (ARM64): https://app.bluebox.ai/download/bluebox?os=linux&arch=arm64
- macOS (Intel): https://app.bluebox.ai/download/bluebox?os=darwin&arch=amd64
- macOS (Apple Silicon): https://app.bluebox.ai/download/bluebox?os=darwin&arch=arm64
- Windows (x86-64): https://app.bluebox.ai/download/bluebox?os=windows&arch=amd64

Steering 2—Production query with Bluebox CLI

Run the command `bluebox skills get bluebox-production-query` to get the
relevant Bluebox skill content. Copy the output directly into the Customer
steering field in Kiro.

3. Create an access token

In Bluebox → Account → Access tokens, select Create token.

4. Add the token to Kiro Web

In Kiro Web → Settings → Agent → Secrets, add a secret:

NameValue
BLUEBOX_TOKEN(token from step 3)

Tokens expire after 30 days. When a token expires, create a new one in Bluebox and update the secret in Kiro Web.