OS AI Agent versions independently of the Collectord agent and the Monitoring apps, using standard semantic versioning.
OS AI Agent is in beta and is not on Splunkbase yet; the package is available on request (see Installation). The beta build reports version 0.9.0, and this page describes it; a later beta counts up from there, and the first Splunkbase release will be 1.0.0.
0.9.0
The beta build.
- Bring your own model. Four provider kinds - OpenAI-compatible (OpenAI and every common local server), Anthropic, Azure OpenAI with key, Entra ID, or managed identity, and AWS Bedrock - with per-provider retry, circuit breaking, and ordered failover. See Providers.
- Chat. Server-side conversations run by a background worker, streaming progress, approval cards for every write, per-conversation provider and approval mode, tool cards with the exact SPL, and export to PDF, web page, Markdown, text, or a signed JSON archive that imports again. See Chat.
- Tools. Data profiling, bounded read-only search, discovery of your macros, lookups, data models, KV Store collections, users, and installed commands, and approval-gated authoring of Dashboard Studio dashboards, saved searches, reports, and alerts, grounded in an embedded copy of the SPL and Dashboard Studio references. See Tools.
| osai. A completion, an SPL explanation, or an unattended read-only investigation from any search or scheduled report. Every run is a chat job: gated and metered like a chat turn, and readable in Chat afterwards. See The osai command.- No process a user controls holds a provider key. The app’s roles do not carry
list_storage_passwords; a key is read for the agent in a privileged context, is only ever sent to the provider it was stored for, and never follows a redirect. The read-only search tools refuse macros andsavedsearch:datasets rather than run them unchecked; the license key, the activity trail and retention need a full Splunk administrator; and model output is rendered through one sanitizer that keeps text, tables, code and links and drops images, forms, media and inline styles. See Security. - Alert investigations. The “Ask OS AI Agent” alert action, running as a named user on a token minted for the job. See Alerts and schedules.
- Query templates and situational guidance for teaching the agent your conventions.
- Cost accounting and spend limits over a rolling 24-hour window, with no bundled price list. See Costs and limits.
- An activity trail in
_internal, and four dashboards: Token Usage, Activity Trail, Health, and a Conversation inspector. See Dashboards. - A Setup page - providers, settings, guidance, query templates, and license - that is the app’s full configuration surface, which is what Splunk Cloud requires.
- Search head clustering support, with configuration replicated and each chat job run by exactly one member.
- Free tier: one provider and one user for non-commercial use, no license key required.
Requires Splunk Enterprise 9.4, 10.0, 10.2, or 10.4 on a search head with KV Store, or Splunk Cloud. See Requirements for the full compatibility matrix.
Known issues and limitations
- Per-component log levels are configuration-file only, so on Splunk Cloud the log level is raised for the whole app rather than for one component.
- Azure managed identity requires an Azure-hosted search head; it does not apply on Splunk Cloud.
- Alert investigations need token authentication, which is off by default on Splunk Enterprise.
- Right after an app upgrade without a splunkd restart,
| osaimay refuse to run and say the REST handlers predate the command; restart splunkd, or wait for the handler processes to recycle.