1. Collect
Content-free findings from the browser, governance events from the SDK and the runtime, scan reports from the scanner, activity from the MCP server.
Observe · Agent Observability
See adoption, cost
and activity in one place.
Every session lands in one audit trail with the person, the agent and the findings attached. Lustena shows where AI is used, how much, by whom, and what it touched, and exports the evidence your auditor asks for.
Works with
ChatGPTClaudeGeminiCopilotClaude CodeCursorCrewAILangGraphYou can't manage what you can't measure
Lustena measures every session.
Without a control plane On the control plane
Adoption
Without a control plane: The first consolidated number about AI use arrives on the license invoice, by team, months late.
On the control plane: Sessions per app, per agent and per team, from the browser and from code, updated as they happen.
Cost
Without a control plane: Tokens burn inside agents nobody declared. Licenses pile up for people who stopped using them.
On the control plane: Sessions per person and app show which paid seats sit idle; model and volume per agent through the SDK and the LiteLLM connector give token spend an owner.
Demand
Without a control plane: Nobody knows which teams already rely on AI, for what, until someone asks for budget.
On the control plane: The GPTs, projects and connectors each team set up on its own show where AI is already in demand, and where an official agent should go next.
Evidence
Without a control plane: Preparing for a questionnaire means a week of screenshots and emails.
On the control plane: An evidence package per agent or per workspace, with inventory, limits, findings and decisions, mapped to the frameworks you name.
Adoption
Sessions per app, per agent and per team
Productionlast 7 days
Adoption and cost
Who uses which agent and how much, with an owner for every line of spend.
1,642
sessions
5
teams
12
idle paid seats
Sessions by team
- Marketing612
- Sales438
- Engineering373
- Finance141
- Support78
Model and volume by agent
| Agent · model | Volume | Spend owner |
|---|---|---|
| Support bot (LangGraph)gpt-4.1-mini · SDKowner · platform team | 18.4M tok | platform team |
| Contract review (CrewAI)claude-sonnet-4.5 · LiteLLMowner · joao@acme.com | 6.2M tok | joao@acme.com |
| Lead enrichmentgemini-2.5-flash · LiteLLMowner · rafa@acme.com | 2.9M tok | rafa@acme.com |
Idle paid seats · no session in 30 days
12 of 60 · ChatGPT Business 9 · Claude Team 3
Observability for agents, not just models
The trail starts at the device and the framework and ends in the tools your security and compliance teams already watch.
2. Correlate
Sessions, runs and parent-child relationships are preserved, so a finding is shown inside the session and the agent that produced it.
3. Report
Agent 360 views, sessions per week, findings by class, limits crossed, devices enrolled and what remains outside coverage.
4. Export
Webhooks for alerts and an evidence package to share with your auditor.
Signals
- Browser findings
- Governance events
- Scan reports
- Device heartbeats
Views
- Agent 360
- Sessions and turns
- Findings inbox
- Coverage and gaps
Audit
- Attributed trail
- Audited content reads
- Evidence package
- Retention controls
Integrations
- Webhooks to Slack and SIEM
- LiteLLM
- OpenTelemetry ingestion
- Evidence package export
Model observability is not agent observability.
Token counts and latency tell you what the model did. They do not tell you which agent acted for which person, what data it touched and whether it stayed inside its limit. Lustena observes the agent.
Outside the loop · where legacy controls sit
- LLM gateway metrics
- APM & tracing
- Cloud billing
- SaaS usage reports
No line of sight inside
Inside the loop · where Lustena sits
Agent
every action →Lustena← result · next step
System
The agent, the person, the session and the finding, correlated.
Questions
What is in the evidence package?
The agent inventory with owners, declared limits and their versions, findings with lifecycle, scan reports and the decisions taken, mapped to the frameworks you select.
Can I send alerts to my SIEM?
Yes, through webhooks. Each alert carries the agent, the person, the dimension crossed and one sentence of explanation, never content.
How long is data kept?
Content-free findings follow your workspace retention. Redacted content, when enabled, has its own retention of up to 365 days and is deleted by a scheduled sweep.