SolarWinds

Agents act inside the customer's self hosted SolarWinds Observability deployment through one connector: Platform Connect and a connection plugin on a separate server that reaches the SolarWinds Information Service API, with every action running under the permissions of the SolarWinds account each user supplies at onboarding.

Through it, goals tune alert thresholds and alert actions, retire stale nodes and run approved remediation scripts on network devices managed by Network Configuration Manager. Connections to other SolarWinds products and to third party systems are planned rather than available, teams cannot add a tool of their own, and Avior is open only to customers admitted to SolarWinds' early access program.

Sourcedocumentation.solarwinds.com/en/success_center/ai/content/ai-onboarding.htmread 2026-11-09

Each goal runs as a control loop of role based agents. They collect the goal's metrics, record findings tagged from Critical to Low, test hypotheses, write proposals that separate agent actions from human tasks, execute what is permitted and then measure the metrics again on the next iteration, reporting whether the outcome was achieved, improved, unchanged or degraded.

A paused goal resumes only once every action group from earlier iterations is approved, declined, voided or running, and sessions chain skills and SolarWinds Query Language queries across stages with a plain language update after each one. The whole loop sits behind an early access program that SolarWinds opens to eligible customers under its own entitlement decisions. Once open to every customer, it would let a team hand a standing outcome to coordinated agents and get a verdict back on every pass.

Sourcedocumentation.solarwinds.com/en/success_center/ai/content/ai-goals.htmread 2026-11-09

Context is gathered from the connected SolarWinds deployment at run time.

During onboarding Avior scans the environment for nodes and their states, firing alerts and compliance violations, and each session or goal then pulls metrics, alerts, configurations, change and incident history and topology through scoped SolarWinds Query Language tools, flagging data that is missing or aged out rather than guessing.

Goals, skills and tools carry SolarWinds' own operational guidance, reasoning steps and guardrails. Tickets and knowledge articles enter only when a connected system supplies them, and today the only connected system is the self hosted observability deployment, all inside an early access program.

Sourcedocumentation.solarwinds.com/en/success_center/ai/content/ai-onboarding.htmread 2026-11-09

Autonomy comes in three levels: Conservative lets only read only actions run unattended, Balanced adds routine nondestructive changes and Autonomous lets everything run. Owners and admins set an organization ceiling no personal setting can exceed, new goals start in Plan Mode, which stops after planning and executes nothing, and proposals reach an Inbox with benefit and risk ratings to approve or decline, beside human tasks the agent hands back.

Only a goal's owner can approve its actions, Members approve nothing, and unknown or unclassified tools fail closed until someone approves them. These controls ship inside an early access program limited to eligible customers. Once open to every customer, those settings would let each organization choose which changes run alone and which wait for a named person.

Sourcedocumentation.solarwinds.com/en/success_center/ai/content/ai-set-up-autonomy.htmread 2026-11-09

Users sign in through the same identity service as SolarWinds Observability SaaS, and an organization can turn on SAML single sign on, require MFA through an authenticator app with a trusted device timeout of 1 to 90 days, and shorten session expiry to between 1 and 30 days. Owner, Admin and Member roles split billing, user management and approvals from daily use, and Avior inherits the rights and account limitations of the SolarWinds account each user connects.

SolarWinds holds ISO/IEC 27001:2022 certification, with 27017 and 27018 controls, for the systems that build and run its products and services; its SOC 2 Type 2 reports cover Observability SaaS, Service Desk, Pingdom and Loggly, not Avior, which remains in early access. Once open to every customer, those controls would tie Avior sign in to the organization's own identity provider.

Sourcedocumentation.solarwinds.com/en/success_center/ai/content/settings/users-swicus.htmread 2026-11-09

Sessions expose each tool call, skill and SolarWinds Query Language query with its result, post a plain language update at every stage and keep generated evidence such as node health snapshots and traffic charts in an artifacts panel, closing with a summary of findings and numbered next steps.

Goal pages hold each iteration's metrics, findings, proposals and actions, the Inbox keeps a History tab of decisions, admins can open an organization activity log that Members cannot, and a usage page tracks LLM usage and credit consumption. Some activity and audit views depend on the release a customer runs, and Avior is open only to early access customers. Once open to every customer, that record would let a team trace any change back to the query, evidence and approval behind it.

Sourcedocumentation.solarwinds.com/en/success_center/ai/content/ai-investigate-issue.htmread 2026-11-09

Each user keeps a personal memory in settings, free text about priorities, the clusters they care about or how much detail they want, which Avior reads beside the verbosity and autonomy choices.

Sessions are saved in the cloud so a conversation resumes with the context already gathered, and each goal keeps an activity history of metrics, findings, proposals and actions that the next iteration measures against.

The personal memory is written by the user, and retained prompts, outputs and logs follow SolarWinds' general privacy and retention policy rather than a lifetime of their own. Avior is available only through early access for eligible customers.

Sourcedocumentation.solarwinds.com/en/success_center/ai/content/ai-personal-settings.htmread 2026-11-09

Avior is a multitenant SaaS application with three entry points, AWS data centers in North America, in Frankfurt for Europe and in Australia, while a connection plugin runs on a customer Windows or Linux server beside the self hosted SolarWinds deployment, reaching it on port 17774 and the cloud on port 443.

During early access, retained prompts, context, outputs and logs are stored in the AWS US East 1 region whichever entry point a customer uses, and matching that storage to the region of the customer's Platform Connect instance is planned for general availability. Avior has no self hosted or private cloud edition.

Sourcedocumentation.solarwinds.com/en/success_center/ai/content/ai-onboarding.htmread 2026-11-09

Nine curated goals ship ready to activate, each a complete operational job with its own metrics: Optimize Alert Thresholds, Reduce Alert Flapping, Reduce Alert Noise, Repair Alert Automation, Cleanup Monitoring Inventory, Prevent Capacity Incidents, Ensure Config Backups, Reduce Compliance Violations and Remediate Config Drift.

Reduce Compliance Violations, for one, audits policy violations, writes and registers remediation scripts for uncovered rules and runs the approved ones toward CIS, NIST and STIG baselines, tracking open violations and script coverage. The catalog targets the network engineer role first and reaches customers through SolarWinds' early access program. Once open to every customer, a team could switch on a ready made operational job without writing its own playbook.

Sourcedocumentation.solarwinds.com/en/success_center/ai/content/ai-goals-overview.htmread 2026-11-09

Goals run on a cadence the team chooses when it defines the goal, so iterations start on schedule with no one prompting them, and under Balanced or Autonomous settings a goal keeps that schedule even while its proposals wait for review.

People start sessions from the prioritized issue list on the home page with Investigate All, from a node's page in the Environment view, from suggested follow up questions or by typing a request, and a Daily Digest email arrives at a time each user picks with priorities, pending approvals and goal status. Avior reaches customers only through an early access program for eligible SolarWinds customers. Once open to every customer, those schedules would keep goals working overnight and on weekends with no one starting a run.

Sourcedocumentation.solarwinds.com/en/success_center/ai/content/ai-goals.htmread 2026-11-09

Avior sends the minimum context a request or goal needs to a managed model provider that SolarWinds selects, by way of the SolarWinds platform rather than straight from the customer's network, and model calls are stateless with no training on customer data. SolarWinds names OpenAI and Microsoft Azure among the companies supplying its language models, and its SaaS assistant routes through an LLM gateway to AWS Bedrock, but Avior names no model for its own agents and has no setting to choose one. Credits meter that model use per question and per goal iteration, inside an early access program.

Sourcep1.aprimocdn.net/solarwinds/4ac713f9-d886-4679-8af9-b4d700ff0caf/frequently_asked_questions_-_solarwinds_avior_data_compliance_original%20file.pdfread 2026-11-09

Avior offers no public API, SDK, MCP server or agent card of its own, so goals and sessions start only inside its web interface. The SolarWinds Information Service API it depends on belongs to the monitored self hosted deployment and is the channel Avior uses to read and change that estate, not a way for outside software to call Avior. Third party systems are slated to connect over the Model Context Protocol with Avior as the caller rather than the system being called, and that connection is not yet available in the early access release.

Sourcedocumentation.solarwinds.com/en/success_center/ai/content/ai_documentation.htmread 2026-11-09

Plan Mode runs a goal on simulated activity and stops in a Plan Ready state, leaving its proposed actions as read only artifacts a team reviews before turning Plan Mode off for that goal.

Once live, every iteration measures the goal's metrics again, such as flapping configurations over 24 hours or devices without a backup in seven days, and reports whether the outcome was achieved, improved, unchanged or degraded, so each change Avior makes is checked against its effect. Scored test cases and offline evaluation suites are absent, and the product is reachable only through early access. Once open to every customer, that loop would give a team a before and after reading on every change, with a dry run first.

Sourcedocumentation.solarwinds.com/en/success_center/ai/content/ai-plan-mode.htmread 2026-11-09

Avior works only through supported product connections and does not log in to or operate arbitrary machines, browsers or desktops. Every read and change goes through the SolarWinds Information Service API of the connected self hosted deployment and, from there, through modules such as Network Configuration Manager to network devices, so a system with no programmatic interface sits outside its reach. Its own web interface, with sessions, goals, the Inbox and the Environment view, is where people direct the agents rather than something the agents drive.

Sourcesolarwinds.com/blog/introducing-avior-goal-oriented-ai-for-it-operationsread 2026-11-09