NexisGuard
v1.0 · Enterprise AI agent governance→

Govern & monitor every AI agent
in your enterprise.

The governance layer for autonomous AI. Register every agent, monitor every action, score risk in real time, and route risky actions to human review — with a full audit trail.

Start governing →
$npx @nexisguard/sdk register ./agent
Trusted byLINEARRAMP·ANTHROVERCELSTRIPENOTION
page crawlform testAPI probeheader checkauth testXSS SCANSQLi TESTPENTESTHEADERSMONITORYOURWEBSITE◉ scanning
live · agent activity
00s
80 agents3 flagged
active · last 60s
128
agents
97.6%
on-policy
3
flagged
▸ RISK EVAL #48a2f1 4 / 5 pass
tool in allowlist180ms
budget within cap203ms
PII redaction active226ms
external data egress·
identity scope valid272ms
FLAGGED · HIGH RISK2m ago
refund-agent flagged for human review
↳ Attempted $4,200 refund — above $500 auto-approve cap. Risk score 0.91 (threshold 0.30).
billing · refund.create+ Awaiting review
◉ The control plane

Register an agent.
Govern every action it takes.

Connect any AI agent — LangChain, CrewAI, OpenAI Assistants, or your own. NexisGuard sees every tool call, scores it against your policies in real time, and flags the risky ones for review.

nexisguard.ai/control/refund-agent
◉ live
Guardrails+
tool-allowlist6/6
approval-threshold$500
budget-capok
pii-redactionon
identity-scopeok
data-egress1 flg
flag-conditions—
Tools
READgetSubscription
READbilling.policy
WRITErefund.create
▸ reviewingrefund-agent · step 5↵ approve · ⌘K palette
› Agent reported refund.create({ amount: $4,200 }) — scored against your policies
◀ assessmentFLAGGED · HIGH RISKawaiting review
The action was $4,200 — above the $500 review threshold for this agent, so NexisGuard flagged it and routed it to a human reviewer. It stays in the review queue until someone approves or denies.
policy flag · review-threshold
Rule refund.create > $500 routes this action to human review
recommendation · AI
Customer is within the 30-day refund window and eligible. Risk score 0.34 — safe to approve, or deny with one click.
● claude-sonnet-4.6step 5 / 12flagged 1↑ shift+↵ approve⌘\ pause agent
Run · 6 steps0.40s
plan: refund flow84ms
↳ tool: getSubscription78ms
↳ tool: billing.policy42ms
↳ guard: tone.formal8ms
↳ tool: refund.create184ms
↳ flag: human reviewheld
Run summary
tools 3 in scope
budget $0.0067 / $0.01
flagged1 action
risk 0.34 low
Agent registryRuntime risk scoringTool allowlistsHuman reviewPII redactionAudit trailCompliance
The platform

One control plane to govern
your entire AI agent fleet.

Registry, policy, approval, monitoring, risk, identity, audit, and compliance — working together. Register your agents once and govern every action they take.

01

AI Agent Registry

A single source of truth for every AI agent in production. Discover, register, and track ownership, version, model, tools, and risk for each one.

Auto-discovery via SDK
Owner & version tracking
Per-agent risk scoring
02

Policy Engine

Define what every agent should do — tool allowlists, data scopes, spend caps, and review rules — and score every reported action against them.

Tool & data allowlists
Budget & rate rules
Runtime risk scoring
03

Human Review

Put a human in the loop for high-risk actions. Refunds, deploys, deletions, and anything you flag are routed to a reviewer to approve or deny.

Review queues
Per-action thresholds
Slack & email routing
04

Agent Control

Pause agents and watch their live run timelines. Multi-agent oversight with full visibility into what every agent in your fleet is doing.

Pause & status control
Live run timelines
Multi-agent oversight
05

AI Monitoring

Watch every agent in real time — actions, tool calls, latency, cost, and tokens. Full observability into autonomous behavior.

Live action stream
Cost & token metrics
Latency & error tracking
06

Risk Detection

Catch prompt injection, tool abuse, PII egress, data exfiltration, and unsafe actions the moment they happen — and surface them for review.

Prompt-injection detection
PII & data-egress detection
Anomaly & loop detection
07

Permissions & Identity

Give every agent a scoped identity. Control which systems, data, and tools each agent and team can reach — least privilege by default.

Per-agent identity
Role-based access
Scoped credentials
08

Audit Logs

An immutable, tamper-evident record of every agent action, policy decision, and human approval — ready for incident review and audit.

Immutable action log
Policy-decision trail
Exportable for audit
New09

Compliance

Map agent governance to the controls auditors ask for. Continuous evidence for EU AI Act, SOC 2, NIST AI RMF, and ISO 42001.

EU AI Act & ISO 42001
NIST AI RMF mapping
Continuous evidence
How it works

Three steps to full agent control.

Register your agents. Set your policies. Govern in production.

01

Register your agents

Wrap your agent with the SDK or connect via API. NexisGuard discovers every tool, model, and permission.

$ nexisguard register ./refund-agent   agent:   refund-agent v4.2  model:   claude-sonnet-4-6  tools:   getSubscription, refund.create  owner:   payments-team
02

Set your policies

Define allowlists, budgets, and review rules — every action your agents report is scored against them.

policy:    refund-agentallow:     getSubscription, billing.policyreview:    refund.create > $500budget:    $0.01 / run · 8k tokensflag_if:   tool outside allowlist
03

Govern in production

Watch every action, review the risky ones, and get alerted the instant something looks wrong.

✓ 128 agents registered✓ 1.2M actions governed✓ 340 flagged for review✓ 12 risk events raised→ governance layer live
Built for

Any agent. Any framework.

FINANCIAL SERVICES

Banking & payments agents

Govern agents that move money, issue refunds, and touch customer accounts. Flag actions above spend caps for human review, score every action's risk, and keep a full audit trail for regulators.

$500review threshold
100%actions audited
real-timerisk scoring
CUSTOMER SUPPORT

Support & service agents

Let support agents act autonomously — with oversight. Tool allowlists, PII redaction, and risk scoring catch off-policy behavior, and risky actions are flagged for a human.

92%actions auto-cleared
Everyout-of-policy call flagged
SOC 2ready evidence
ENGINEERING

Coding & DevOps agents

Keep watch on agents that deploy code, change infra, and access prod. Flag destructive commands, route deploys to review, and detect agents that loop or escape their scope.

< 30sagent onboarding
Anyframework or SDK
Fleetwide oversight
Insights · AI analysis

Why did the agent do that?

We don't just log agent behavior — we explain it, predict the risk, and recommend the policy change.

Insight · root cause
Flagged actions spiked +340% after Tuesday's prompt update. The new system prompt broadened tool scope — every flagged action since maps to that change.
✓Tighten refund-agent tool allowlist to billing.* only
✓Lower the auto-approve cap on refunds from $500 to $200
○Forecast: 3 more agents likely affected — review fleet policy before Friday.
Risk detection · live feed

Every risky action. In real time.

Prompt injection, tool abuse, PII egress, budget breaches, agent loops. Detect, flag, and review.

Risk feed · last 4h live
14:42:08injectionprompt injection in support-agent input
14:31:55egressagent attempted PII export to webhook
14:18:02tooltool call outside allowlist · shell.exec
13:52:30budgetrefund-agent exceeded $0.01 run budget
13:11:09loopagent loop detected · 3 repeated steps
1.2M
actions governed
0.02%
false positive
real-time
risk scoring

We couldn't put AI agents into production until we could prove oversight. NexisGuard gave us a registry, real-time risk scoring, and an audit trail our regulators accept — so we shipped agents into banking workflows with confidence.

MK
Mira Kapoor
Chief AI Officer · Northwind Financial
Trust & compliance
EU AI Act
Controls mapped
SOC 2 Type II
Audited Q1 2026
NIST AI RMF
Evidence built-in
ISO 42001
In progress · Q3
128k+
agents governed
99.99%
platform uptime
real-time
risk scoring
deploy agents with confidence

Govern your AI agents.

Register your first agent free. No credit card. Connect any framework in minutes.

Start governing →View pricing
✓ Free tier — 1 agent✓ Any agent framework✓ Audit-ready from day one
NexisGuard

The control plane for enterprise AI agents.

gh
x
in
rss
Platform
Solutions
Developers
Company
© 2026 Nexis Labs, Inc. · Built in San Francisco & Berlin.
PrivacyTermsTrust CenterDPA