L0
Inventoried
14-day trial
TBDto be defined
The unit exists in the inventory, with owner and group. Manual data.
Requirement
None
- Registry and taxonomy
- Human owner per unit and group
- Role documentation
- Navigable fleet map
Pricing
Start by inventorying. Raise the level at your pace.
Four integration levels, all with a 14-day trial. Pricing is TBD while we finalize the model — tell us about your fleet and we'll show you the exact path.
L1Entry point
Observed
Incoming telemetry · 14-day trial
TBDto be defined
Activity, results and costs flow into Kordana. Health and alerts appear.
Requirement
OTel GenAI · lightweight SDK · webhook
- Everything in L0
- Operational status
- Business metrics by unit and group
- Alerts routed through the owner chain
- Basic scorecards
L2
Controlled
Operational control · 14-day trial
TBDto be defined
Command channel: stop/play, limits, structured handoff, inline approval.
Requirement
Control hooks (SDK or gateway)
- Everything in L1
- Stop / Play / Freeze
- Degraded mode
- Escalation inbox
- Blocking approvals
L3
Managed
Full lifecycle · 14-day trial
TBDto be defined
Full lifecycle from Kordana: design, probation, versioning, verified retirement.
Requirement
Deploy via Kordana or deep integration
- Everything in L2
- Role editor with blocking KPIs
- Automated probation
- One-step versioning and rollback
- Retirement with a certificate
02·Capability comparison
Each level adds a maturity layer.
Move up when operations justify it. You don't pay for what you aren't using yet.
| Capability | L0 | L1 | L2 | L3 |
|---|---|---|---|---|
AI unit inventory Registry, taxonomy, owner and group. | ||||
Navigable fleet map Hierarchical view of roles, groups and dependencies. | ||||
Telemetry and operational status Activity, costs, results and health in real time. | ||||
Business metrics and scorecards KPIs per unit, group and fleet with routed alerts. | ||||
Stop / Play / Freeze Direct control commands on units. | ||||
Structured handoff and approvals Human escalation with context and inline approval. | ||||
Role editor with blocking KPIs Define the role and success thresholds before operating. | ||||
Automated probation Performance validation before promoting to production. | ||||
One-step versioning and rollback Versioned, reversible role changes. | ||||
Retirement with a certificate Token/permission wipe and retirement evidence. | ||||
Multi-framework LangGraph, CrewAI, Bedrock, Vertex, OpenAI, your own code. | ||||
Model gateway + BYOAI Bring your keys. Tier routing. Token accounting. | ||||
A/B/C buckets with deny-list Judges only on kordana-hosted. Fail-closed on doubt. | ||||
Edge enforcement (PDP → PEP) Ed25519-signed policy enforced at your gateway. | ||||
Evals flywheel Structured feedback → versioned bank → certification. | ||||
Granular IAM + dual-layer RLS Roles per subtree, default-deny, no service_role. | ||||
Work windows Shifts, scheduled freezes and off-hours regimes. |
Not includedIncluded
03·First-class multi-framework
Kordana doesn't dictate where the agent was built.
No entity in the model assumes the unit was built here. The Kordana base app is one origin among many.
- LangGraph
- CrewAI
- Vertex AI Agent Builder
- AWS Bedrock Agents
- Copilot Studio
- OpenAI Agents
- Your own code
- Kordana base app
04·Frequently asked
What the first 20 conversations ask.
Do I have to migrate my agents to Kordana?
No. Kordana is first-class multi-framework. The Kordana base app is one origin among several; most customers start with agents already running in another framework.
What do I need to reach L1 (observed)?
One of three paths: OpenTelemetry GenAI, our lightweight SDK, or a simple webhook that reports activity, results and costs. From there, metrics, alerts and scorecards appear.
And for L2 (controlled)?
Control hooks exposed via SDK or gateway. That enables stop/play, live limits, structured handoff and inline approval.
What about agents built with the Kordana base app?
They're born in L3 — full lifecycle managed from Kordana — with no additional work.
Can I mix levels in the same fleet?
Yes. Every unit shows its level in the UI at all times, with a suggested upgrade path (what's needed to level up). The level is also the account-expansion ladder.
How does BYOAI work?
Bring your keys for OpenAI, Anthropic, Google, Azure, Bedrock or Vertex — or even your own gateway. Kordana encrypts them per workspace, resolves credentials just in time and routes by quality tier. The gateway never accepts a client-supplied model name: it routes by tier.
Does Kordana sit in the data path of my models?
No. For clients with their own gateway, Kordana acts as decision plane (PDP/PAP) and pushes signed policy to the PEP in your gateway. We reconcile via attestations and echoes, but we never see inference traffic.
Request a personalized demo.
Every level starts with a 14-day trial. Pricing is TBD — tell us about your fleet and we'll show you the exact path.