Free
100k
governed events / month
$0
See it · 5 agents · 2 seats · 30-day trail visibility
Full console, policy, approvals, trail API
Start free
Pricing
How much agent activity do you govern?
A governed event is one tool call your agents make that Shield allows, reviews, or denies - recorded on your trail. A governed agent is an AI agent you register with Shield - one Claude Code, Cursor, or custom bot acting on your systems. Events are metered, never dropped. Console seats are humans who sign in.
Starter
500k
governed events / month
$29 / month
Run it · 25 agents · 5 seats · 30-day trail visibility
One team's agents in production
Get startedPro
5M
governed events / month
$149 / month
Prove it · 250 agents · 25 seats · 90-day trail visibility
Signed evidence bundles · claim verify API
Get startedScale
custom
governed events / month
$499 / month
Prove it org-wide · unlimited agents + seats · 365-day trail
SSO / SCIM · Gatekeeper included · Scale support
Get startedFree tier, side by side
| Shield KYA Free | Typical agent-observability free tier | |
|---|---|---|
| Monthly free volume | 100k governed events - metered, never dropped | 100k spans - hard cap |
| Retention | 30-day trail visibility | 14 days |
| When it acts | Before the tool call runs - policy gate + human approvals, blocking built in | After the fact - flagging only, no blocking |
| Included on free | Policy gate, approvals, agent passports, Shield Verified badges, 5 agents, 2 seats | Tracing and dashboards |
Observability tools and KYA do different jobs: they explain what happened, KYA decides what may happen and proves it. Most teams run both.
Plan details
| Feature | Free · $0 | Starter · $29/mo | Pro · $149/mo | Scale · $499/mo |
|---|---|---|---|---|
| Usage & limits | ||||
|
Governed events per month
One tool call your agents make that Shield allows, reviews, or denies -
recorded on your trail. Metered, never dropped: at the quota you get a
warning - what happens past it is in the Overage row below.
|
100,000 | 500,000 | 5,000,000 | Custom |
|
Overage
What happens past the monthly event quota. Events are never dropped.
|
Upgrade for more | Metered, billed monthly - events never dropped | Metered, billed monthly - events never dropped | Custom |
|
Governed agents
An AI agent you register with Shield - one Claude Code, Cursor, or
custom bot acting on your systems. Distinct identities (ACTIVE or
SUSPENDED) at one time. Kill or revoke frees a slot.
|
5 at a time | 25 at a time | 250 at a time | Unlimited |
|
Console seats
Humans who sign in to approve and operate. Not company headcount.
Revoking an invite or removing a member frees a seat.
|
2 | 5 | 25 | Unlimited |
|
Trail visibility
How far back you can read the trail - dashboard windows, trail API,
session trails. Visibility only: nothing is ever deleted, and
upgrading shows further back immediately.
|
30 days | 30 days | 90 days | 365 days (custom on request) |
| Identity & registry | ||||
| Agent registry (register, list, status) | Yes | Yes | Yes | Yes |
| Agent passport (HTML + JSON download) | Yes | Yes | Yes | Yes |
| Policy-gated spawn (agent creates agent) | Yes | Yes | Yes | Yes |
| Human mint mode (allow / break-glass / approve) | Yes | Yes | Yes | Yes |
| Session clearance (Read / Build / Deploy) | Yes | Yes | Yes | Yes |
| HIGH-risk shrink mode (operator / auto / approve) | Yes | Yes | Yes | Yes |
| Signed agent passport + session claim | Yes | Yes | Yes | Yes |
| Verify claim API | No | No | Yes | Yes |
| Orphan / parent-tree report | No | No | Yes | Yes |
| Version hash on register (prompt + tools + policy) | Yes | Yes | Yes | Yes |
| Kill / revoke agent | Yes | Yes | Yes | Yes |
| Policy & human approval | ||||
| Runtime policy: ALLOW / DENY / REQUIRE_APPROVE | Yes | Yes | Yes | Yes |
| Fail-closed defaults (missing approval → deny) | Yes | Yes | Yes | Yes |
| Policy playground (console evaluate) | Yes | Yes | Yes | Yes |
| Policy evaluate API (MCP / OpenAPI / HTTP) | Yes | Yes | Yes | Yes |
| Approvals queue (approve / reject) | Yes | Yes | Yes | Yes |
|
Approval priority queue
Escalate or de-escalate Holds so the riskiest tool call is answered
first. HIGH-risk sessions auto-prioritize on every plan.
|
No | No | Yes | Yes |
| PII redaction before model calls | Yes | Yes | Yes | Yes |
| Observe, evidence & trail | ||||
| Session ingest + session list | Yes | Yes | Yes | Yes |
| Session timeline (tools, risk, raw payload) | Yes | Yes | Yes | Yes |
| Session risk only raises severity (never auto-allows) | Yes | Yes | Yes | Yes |
| Evidence & cases | Yes | Yes | Yes | Yes |
| Hash-chained event / audit trail | Yes | Yes | Yes | Yes |
|
Shield Verified badge endpoints
Public SVG / JSON on every plan - anyone can check a badge.
|
Yes | Yes | Yes | Yes |
| Case export (JSON) | Yes | Yes | Yes | Yes |
| Evidence graph dual-write (audit module) | Yes | Yes | Yes | Yes |
| KYA metrics (console) | Yes | Yes | Yes | Yes |
| KYA metrics JSON API | Yes | Yes | Yes | Yes |
|
Natural-language activity search
Type "dangerous" or "what did the agent change" and get matching tool
calls. BM25 fused with local semantic embeddings, ranked by relevance.
|
Yes | Yes | Yes | Yes |
|
Analytics dashboard
Verdict mix, modes, projects, top sessions, and activity volume over
the trail window.
|
Yes | Yes | Yes | Yes |
|
Per-session change diffs
What actually changed per file in a session, not just what ran.
|
Yes | Yes | Yes | Yes |
|
Shared reports (public links)
Token-gated, rate-limited read-only report links; can embed certify
status for auditors.
|
Yes | Yes | Yes | Yes |
|
Trail & cases CSV export
Export the governed trail and case snapshots for your own warehouse
or auditor.
|
No | No | Yes | Yes |
|
Disputed agent cases (console)
Open a case on a disputed agent action, attach evidence, track the
investigation.
|
Yes | Yes | Yes | Yes |
| Dispute create & status-update API | No | No | Yes | Yes |
|
Signed evidence bundle verify (public API)
Anyone can verify a kya certify bundle against the public endpoint,
no account needed.
|
Yes | Yes | Yes | Yes |
|
OTLP span intake
Ingest OpenTelemetry spans from the local gate as governed trail
events.
|
Yes | Yes | Yes | Yes |
| Gateway & MCP | ||||
|
Hosted MCP gateway console
Listeners, routes, backends, policies, and a dry-run playground for
every MCP server your agents call.
|
Yes | Yes | Yes | Yes |
|
Tool & MCP inventory
Score and review every tool and MCP server your agents can reach.
|
Yes | Yes | Yes | Yes |
| Import, investigation & certify | ||||
|
Trace import (LangSmith / Langfuse / Phoenix / OTLP)
Upload a trace export and replay it into the governed trail:
/app/kya/import or POST /api/v1/kya/import. Monthly imported-event
allowance per plan; imported events meter like any governed event.
|
10k / month | 100k / month | 500k / month | Unlimited |
|
Incident detection (deterministic detectors)
Six detectors scan the trail for PII leaks, deny spikes, approval
loops, never-list repeats, unknown-tool holds, and slow calls.
Manual run on every plan (Free: 7-day window, 1 run/hour).
|
Manual · 7-day | Daily · 30-day | Hourly · 90-day | Hourly · 90-day |
|
Fix briefs per incident
What happened, the suggested policy or config change, and the commands
to verify the fix.
|
No | Yes | Yes | Yes |
|
Incident webhook alerts
kya.incident.opened delivered to your HTTPS endpoints with backoff
retry when a scan opens a critical or high incident.
|
No | No | Yes | Yes |
|
Hosted certify gap report
The 30-requirement Agent Trust Baseline (EU AI Act crosswalk) evaluated
live against your hosted trail.
|
Yes | Yes | Yes | Yes |
|
Hosted signed evidence bundle
ed25519-signed bundle your auditors verify against the public endpoint,
no account needed on their side.
|
No | No | Yes | Yes |
|
Continuous trace sync (pull connectors)
LangSmith and Langfuse connectors poll your projects every 15 minutes
and stream new runs into the governed trail, same mapper and quota as
upload import. Credentials encrypted at rest; manage at
/app/kya/connectors.
|
No | No | No | Yes |
| Edge & complements | ||||
| Edge status page (Gatekeeper complement flags) | Yes | Yes | Yes | Yes |
| Shield Agent Gatekeeper (edge after Shield approves) | No | No | Optional | Included / assisted |
| Connectors & install | ||||
| Marketplace install (Claude, ChatGPT, Grok, Gemini, AWS, Azure, GCP) | Yes | Yes | Yes | Yes |
| MCP connector surface | Yes | Yes | Yes | Yes |
|
Outbound webhooks (approval alerts)
POST approval-pending events to your HTTPS endpoint with backoff
retry. Route alerts into Slack, PagerDuty, or your own queue.
|
Yes | Yes | Yes | Yes |
| OpenAPI / GPT Actions surface | Yes | Yes | Yes | Yes |
| Install agents hub in console | Yes | Yes | Yes | Yes |
| Console, tenant & access | ||||
| Hosted operator console (dashboard + all KYA pages) | Yes | Yes | Yes | Yes |
| Tutorial mode + welcome tour | Yes | Yes | Yes | Yes |
| Tenant isolation | Yes | Yes | Yes | Yes |
|
Roles: OWNER / ADMIN / OPERATOR / READ_ONLY
OWNER manages SSO. OPERATOR can approve Holds; READ_ONLY cannot.
|
Yes | Yes | Yes | Yes |
| API keys (machine access) | Yes | Yes | Yes | Yes |
|
SSO / SAML / OIDC login
One IdP per workspace. Operators only - agents keep API keys.
|
No | No | No | Yes |
| SCIM 2.0 user & group sync | No | No | No | Yes |
| IdP group → role maps (group RBAC) | No | No | No | Yes |
| Profile, security, billing settings | Yes | Yes | Yes | Yes |
| Support & onboarding | ||||
| Email support | Community (docs + GitHub) | Standard | Priority | Priority |
| Slack or dedicated channel | No | No | No | Yes |
|
Severity response targets
Product policy for first response - not credits or indemnity.
See Scale support.
|
No | No | No | Yes · Sev-1–4 |
| Guided setup / onboarding help | No | No | No | Yes |
| Local free for development (self-host, see repo docs) | Included for every plan. No seat charge on local-only runs | |||
Prefer to try first? Create a free console account or sign in. Local development stays free (see repo docs). Marketplace install is on every plan. Install hub.