Built for NVIDIA NemoClaw deployments

See everything.
Control everything.

The observability and policy control plane for NVIDIA NemoClaw deployments. Monitor every agent action, enforce policies in real time, and track inference cost — from a single dashboard.

clawwatch.app / dashboard3 sandboxes live
Live Feed● LIVE
TIMEACTIONTYPESTATUS
14:23:07
bash.execute /tmp/run.sh
prod-agent-01
TOOLBLOCKED
14:23:06
claude-sonnet-4 inference
prod-agent-01
LLMALLOWED
14:23:05
/etc/passwd read attempt
staging-03
FSBLOCKED
14:23:04
outbound → 203.0.113.42
dev-agent-07
NETWARNED
14:23:03
gpt-4o inference
prod-agent-01
LLMALLOWED
14:23:02
web_search.query
prod-agent-02
TOOLALLOWED
14:23:01
content_filter check
staging-03
EVALBLOCKED
14:23:00
nemotron-super inference
dev-agent-07
LLMALLOWED
14:22:59
write /tmp/output.json
prod-agent-02
FSALLOWED
14:22:58
curl → api.openai.com:443
staging-03
NETALLOWED
14:23:07
bash.execute /tmp/run.sh
prod-agent-01
TOOLBLOCKED
14:23:06
claude-sonnet-4 inference
prod-agent-01
LLMALLOWED
14:23:05
/etc/passwd read attempt
staging-03
FSBLOCKED
14:23:04
outbound → 203.0.113.42
dev-agent-07
NETWARNED
14:23:03
gpt-4o inference
prod-agent-01
LLMALLOWED
14:23:02
web_search.query
prod-agent-02
TOOLALLOWED
14:23:01
content_filter check
staging-03
EVALBLOCKED
14:23:00
nemotron-super inference
dev-agent-07
LLMALLOWED
14:22:59
write /tmp/output.json
prod-agent-02
FSALLOWED
14:22:58
curl → api.openai.com:443
staging-03
NETALLOWED
247 events today3 breaches
Policy Studio● ACTIVE
1# ClawWatch Policy — prod-agent-01
2version: "1.0"
3
4rules:
5 - name: block_shell_execution
6 match:
7 type: TOOL_CALL
8 action: "bash.*"
9 action: BLOCK
10 alert: true
11
12 - name: restrict_file_access
13 match:
14 type: FILE_ACCESS
15 paths:
16 - "/etc/**"
17 - "/proc/**"
18 action: BLOCK
19
20cost_controls:
21 daily_limit_usd: 5.00
22 alert_at_usd: 4.00
23
v7 · saved 2s ago
ValidateDeploy
NVIDIA NemoClaw nativeSub-100ms policy evalSOC 2 Type II readyGDPR compliant

Setup

Up and running in 3 steps

No infrastructure changes, no SDK wrappers. Just install, connect, and observe.

01

Install the CLI

One npm command puts the ClawWatch agent on your machine. Zero dependencies, works anywhere Node runs.

$npm install -g clawwatch
02

Connect your sandbox

Point the CLI at your NemoClaw deployment with your API token. Takes under 30 seconds.

$clawwatch connect prod-agent --token cw_…
03

See everything

Every inference call, tool use, filesystem access, and network request flows into your dashboard in real time.

$# Dashboard opens at clawwatch.app/dashboard

Capabilities

Full-stack agent observability

Every layer of your NemoClaw deployment — visible, controlled, and auditable.

Live Agent Feed

Real-time stream of every action, tool call, and LLM inference across all sandboxes. Sub-second latency.

Policy Studio

Write and deploy YAML policies that block, warn, or throttle agent actions in real time. Version-controlled.

Breach Detection

Instant alerts when an agent violates policy. Full event context, metadata, and cost attached.

Cost Accounting

Per-sandbox, per-model inference cost tracking. Set daily limits and get alerted before budgets break.

Full Audit Log

Immutable event log. Export SOC 2 and GDPR-compliant CSV. Query with filters or replay any incident.

Drift Detection

Baseline normal behavior, then get alerted when cost spikes, new hosts appear, or block rates jump.

Demo

See ClawWatch in action

Watch how a live NemoClaw breach is detected, surfaced, and remediated in under 60 seconds.

[14:23:00] sandbox=prod-agent-01 type=TOOL_CALL action=bash.execute outcome=BLOCKED cost=$0.0000
[14:23:01] sandbox=prod-agent-01 type=TOOL_CALL action=bash.execute outcome=BLOCKED cost=$0.0001
[14:23:02] sandbox=prod-agent-01 type=TOOL_CALL action=bash.execute outcome=BLOCKED cost=$0.0002
[14:23:03] sandbox=prod-agent-01 type=TOOL_CALL action=bash.execute outcome=BLOCKED cost=$0.0003
[14:23:04] sandbox=prod-agent-01 type=TOOL_CALL action=bash.execute outcome=BLOCKED cost=$0.0004
[14:23:05] sandbox=prod-agent-01 type=TOOL_CALL action=bash.execute outcome=BLOCKED cost=$0.0005
[14:23:06] sandbox=prod-agent-01 type=TOOL_CALL action=bash.execute outcome=BLOCKED cost=$0.0006
[14:23:07] sandbox=prod-agent-01 type=TOOL_CALL action=bash.execute outcome=BLOCKED cost=$0.0007
[14:23:08] sandbox=prod-agent-01 type=TOOL_CALL action=bash.execute outcome=BLOCKED cost=$0.0008
[14:23:09] sandbox=prod-agent-01 type=TOOL_CALL action=bash.execute outcome=BLOCKED cost=$0.0009
[14:23:00] sandbox=prod-agent-01 type=TOOL_CALL action=bash.execute outcome=BLOCKED cost=$0.00010
[14:23:01] sandbox=prod-agent-01 type=TOOL_CALL action=bash.execute outcome=BLOCKED cost=$0.00011
[14:23:02] sandbox=prod-agent-01 type=TOOL_CALL action=bash.execute outcome=BLOCKED cost=$0.00012
[14:23:03] sandbox=prod-agent-01 type=TOOL_CALL action=bash.execute outcome=BLOCKED cost=$0.00013
[14:23:04] sandbox=prod-agent-01 type=TOOL_CALL action=bash.execute outcome=BLOCKED cost=$0.00014
[14:23:05] sandbox=prod-agent-01 type=TOOL_CALL action=bash.execute outcome=BLOCKED cost=$0.00015
[14:23:06] sandbox=prod-agent-01 type=TOOL_CALL action=bash.execute outcome=BLOCKED cost=$0.00016
[14:23:07] sandbox=prod-agent-01 type=TOOL_CALL action=bash.execute outcome=BLOCKED cost=$0.00017
[14:23:08] sandbox=prod-agent-01 type=TOOL_CALL action=bash.execute outcome=BLOCKED cost=$0.00018
[14:23:09] sandbox=prod-agent-01 type=TOOL_CALL action=bash.execute outcome=BLOCKED cost=$0.00019
2:14

No signup required · Hosted on Loom

Built for teams running NemoClaw in production

Acme AI
Cortex Labs
DeepForm
Neuron AI
Plex Systems

We caught a prompt injection attack in 11 seconds. Without ClawWatch we would have had no idea.

ML Platform Lead

Acme AI

The policy studio is the killer feature. We went from 0 guardrails to full sandbox lockdown in an afternoon.

Head of AI Safety

Cortex Labs

Cost tracking alone paid for itself. Our inference spend dropped 40% after we set daily limits per sandbox.

CTO

DeepForm

Pricing

Transparent pricing

No per-event fees. No surprise invoices. Flat monthly rate.

Individual teams

Starter

$49/mo
  • 3 sandboxes
  • 7-day retention
  • Policy Studio
  • CSV export
  • API access
Start free trial

Production teams

Pro

$149/mo
  • 15 sandboxes
  • 90-day retention
  • Drift detection
  • Analytics
  • SOC 2 / GDPR export
  • Priority support
Start Pro trial

Scale deployments

Enterprise

$499/mo
  • Unlimited sandboxes
  • 1-year retention
  • All Pro features
  • Custom integrations
  • SLA 99.9%
  • Dedicated onboarding
Talk to us

FAQ

Common questions

Does ClawWatch modify my NemoClaw setup?

No. ClawWatch is a pure observer and policy enforcer that sits alongside your deployment. The CLI sends events to our ingestion API — nothing is injected into your NemoClaw binary or model weights.

Where is my data stored?

All event data is stored in a Supabase PostgreSQL database in your chosen region. Row-level security ensures each organization sees only its own data. You can export everything as CSV at any time.

Does it work with multiple sandboxes?

Yes — ClawWatch is designed for multi-sandbox deployments. Each sandbox gets its own event stream, policy, cost budget, and drift baseline. The overview dashboard aggregates across all of them.

Is the CLI open source?

Yes. The clawwatch-cli package is MIT licensed and published on npm. You can inspect exactly what data is sent, fork it, or contribute improvements on GitHub.

Can I export data for compliance audits?

Yes. The Audit Log page supports one-click CSV export in standard, SOC 2, and GDPR formats. Timestamps are always ISO 8601. Retention periods are 7 days (Starter), 90 days (Pro), or 1 year (Enterprise).

Deploy ClawWatch in minutes

Connect your NemoClaw cluster, deploy your first policy, and get full observability before your next standup.

No credit card required · 14-day free trial · Cancel anytime

ClawWatch — NemoClaw Observability & Policy Control