Services · Managed Service
Learn your traffic. Audit your agents. Enforce with evidence.
The Starseer Managed Service maps your agents, models, and identities, validates detections against real traffic, and deploys only proven rules. We continuously refine protections on your schedule.
How every deployment runs
Nothing blocks until your traffic proves it should.
01 · Learn
Inventory workflows, models, MCP servers, and identities. Scan every model in the inventory.
02 · Observe
Detections and activation probes run on live traffic and record what they would have blocked.
03 · Review
Walk the results with your team and mark each rule to block, tune, or retire.
04 · Enforce
Promote approved rules to blocking, with a snapshot to roll back to.
Every rule
Runs in observe mode, then earns one of three outcomes.
Block
Promoted to enforcement at the AI Gateway, with a rollback snapshot.
Tune ↺
Adjusted and sent back to observe for another cycle.
Retire
Removed. It never fired on anything that mattered.
AIDR agentic observers
Every session watched. Every finding ready for an analyst.
Agentic observers are the AIDR feature that works between human reviews. They read each session across all three layers, triage what they find against your rules and probes, and hand analysts an assembled case instead of a raw alert.
01 · Watch
Every session through the AI Gateway or AIDR passive mode, including MCP tool calls. 24x7, every tier.
02 · Triage
Score each event against promoted rules, observe-mode rules, and activation probes. Correlate by identity and workflow.
03 · Assemble
Build one case file per finding, mapped to MITRE ATLAS, and route it to Slack or email.
04 · Act
Take the action your tier authorizes, then queue the case for the next human review.
What observers read · Behavioral monitoring reaches the first two layers
Layer 1
Traffic and prompts
What went in and what came out of every model call.
Layer 2
Actions, MCP, and lineage
Every tool call and MCP request, with the identity and workflow behind it.
Layer 3
Inside the model
Activation analysis examines the model's internal computations to understand what drove its behavior and decisions.
Where Starseer excelsThe case file puts the prompt, tool call, MCP request, application trace, and activation probe on one timeline before your analyst opens it.
MCP coverage
MCP is in scope from the first week.
Every MCP server your agents call is inventoried, inspected, and placed under gateway policy, keyed to the identity making the call.
Learn
Inventory
Every MCP server, the tools it exposes, and the identities that call it, mapped per workflow.
Observe
Inspect
AIDR reads MCP tool arguments and results. Observers triage them like any other event.
Enforce
Control
AI Gateway policy applies to MCP tool calls, keyed to the identity making the call.
Connect
Skills and MCP bundle
Starseer skills and MCP configuration route agent tool calls through Starseer. No application code changes.
Managed service tiers
Observers on every session. Analysts on your cadence.
Each tier starts with an onboarding project, then moves into ongoing review. AIDR agentic observers triage every session around the clock on every tier. Starseer analysts review traffic, tune detections, and promote new rules at the tier's cadence.
Monthly review · Advisory
Audit
Setup, config testing, and onboarding, then monthly review. Advisory.
Weekly review · You approve blocks
Monitor
Setup, config testing, and onboarding, then weekly review. You approve blocks.
Daily review · Pre-authorized blocking
Manage
Setup, config testing, and onboarding, then daily review. Pre-authorized blocking.
| What each tier includes | Audit | Monitor | Manage |
|---|---|---|---|
| Onboarding | |||
| Onboarding | Included | Included | Included |
| Discovery, with model scan | Light | Full | Full |
| Workflows secured | 1 | 2 | 2 |
| Custom activation probes | Not included | 2 | 2 |
| Agent identity and attribution | Not included | Included | Included |
| MCP server inventory and coverage | Included | Included | Included |
| Skills and MCP bundle | Included | Included | Included |
| Typical duration | 3 weeks | 4 to 5 weeks | 4 to 5 weeks |
| Ongoing | |||
| Human review | Monthly | Weekly | Daily, business days |
| AIDR agentic observers, immediate alerts | 24x7 | 24x7 | 24x7 |
| Observer action | Alert with case file | Alert, stage candidate rule | Contain in scope, notify |
| Blocking authority | Advisory | You approve | Pre-authorized in scope |
| Reporting | Monthly report | Report + call | Weekly + exec readout |
Launch Readiness upgrade
For any tier with a launch ahead. Extends setup, config testing, and onboarding with a second observe cycle, a purple team assessment of your launch workflows, a partner-facing security summary, and daily review during launch week. 8 to 10 weeks.
What counts as a workflow
One agent or automation, the identities it runs under, and the tools and MCP servers it calls. A different set of identities or tools counts as a second workflow.
How you buy
One license. One managed tier that covers onboarding and after.
One annual subscription, priced in bands by the identities we watch. Bands keep the bill predictable as your agent fleet grows. Moving an agent off a person's account onto its own identity almost never changes your band, so good identity hygiene never costs you more.
Included
Platform license
Starseer Platform License, Small Business, with Guided Start.
Included
Onboarding
Initial setup, config testing, and onboarding.
Included
Managed service
Your tier's review cadence and blocking authority.
Included
24x7 observers
AIDR agentic observers alert immediately. Human response times per your master service agreement.
Monitored identity
Any principal, human or not, sending traffic through the AI Gateway or AIDR in a billing month. Dormant accounts don't count.
Ephemeral roll-up
Sub-agents count as the agent definition they run under, so spawning never inflates the count.
Fair-use volume
Each band includes a pooled session allowance for unusually heavy traffic.
Inside the model
See your AI the way Starseer does.
Activation probes train on your own traffic and read what the model computes, not just what goes in and out.