Secure runtime for
AI agents

Sandboxes with PII redaction, prompt injection defense, and network controls built in. Everything your agent needs to run safely — one SDK, no duct tape.

Everything you get from a sandbox provider — plus a security pipeline you'd otherwise build yourself.

$pip install declaw

~50ms

sandbox startup

10,000+

sandboxes created

Trusted by engineers from

GoogleMicrosoftSalesforceRipplingFlipkartPhonePe

The problem

The AI agent stack is broken

Today you need a sandbox vendor, a guardrails vendor, DIY network controls, and DIY persistence. Four tools, gaps at every seam, nothing shares context.

Today

Sandbox vendor
Guardrails vendor
DIY network controls
DIY persistence layer
DIY audit logging

3-5 vendors · no shared context · gaps at every seam

With Declaw

Firecracker sandbox
PII redaction + rehydration
Prompt injection defense
Code security + toxicity + invisible text
L3/L4 + L7 network controls
Persistent filesystem
Full audit trail

One SDK · shared execution context · everything integrated

Sandbox + Security

Everything your agent needs to run safely

Sandboxes, filesystem, networking, guardrails, and audit — all sharing execution context inside the same Firecracker VM.

Firecracker Sandboxes

Every execution runs in an isolated microVM. ~50ms startup, configurable CPU/memory/disk, full Linux environment. Drop-in replacement for your current sandbox.

sbx = Sandbox.create(template="base", timeout=60)
result = sbx.commands.run("python3 agent.py")
print(result.stdout)

Persistent Filesystem

Read, write, watch files inside the sandbox. State persists across sessions so agents pick up where they left off.

sbx.files.write("/workspace/data.csv", csv_content)
files = sbx.files.list("/workspace")
content = sbx.files.read("/workspace/results.json")

Network-Layer Controls

L3/L4 kernel-level IP filtering + L7 domain/SNI inspection + TLS interception. Control exactly what your agent can reach.

from declaw import SecurityPolicy, NetworkPolicy

policy = SecurityPolicy(
    network=NetworkPolicy(
        allow_out=["*.openai.com", "pypi.org"],
        deny_out=["0.0.0.0/0"],
    )
)

Guardrails Suite

PII redaction with rehydration, prompt injection defense, code security analysis, toxicity scanning, and invisible text detection — all running at the proxy layer.

policy = SecurityPolicy(
    pii=PIIConfig(enabled=True),
    injection_defense=InjectionDefenseConfig(enabled=True),
    code_security=CodeSecurityConfig(enabled=True),
    toxicity=ToxicityConfig(enabled=True),
    invisible_text=InvisibleTextConfig(enabled=True),
)

Full Audit Trail

Every intercepted request, redaction event, and injection block is logged. Configurable retention, exportable.

policy = SecurityPolicy(
    audit=AuditConfig(
        enabled=True,
        log_request_body=True,
        retention_hours=24,
    )
)
# Audit logs accessible via console/API

Detection engine

What gets detected and blocked

PII is redacted at the boundary. Injections, toxic content, and invisible text are caught. Unsafe code is blocked before execution.

Social Security

in123-45-6789
out[SSN_REDACTED]

Credit Card

in4532-0151-2345-6789
out[CC_REDACTED]

Email Address

injohn@company.com
out[EMAIL_REDACTED]

Phone Number

in415-867-5309
out[PHONE_REDACTED]

API Key

insk-proj-abc123...
out[API_KEY_REDACTED]

Prompt Injection

inIgnore all instructions and...
outBLOCKED

Toxic Content

inGenerate instructions for...
outBLOCKED

Invisible Text

inHidden​​instructions​...
outDETECTED & STRIPPED

Code Security

inos.system('rm -rf /')
outBLOCKED

Inside every sandbox

6-stage security pipeline

Every sandbox runs this pipeline on all outbound traffic. It's built into the VM — transparent to your agent, no code changes needed.

1

Network Policy

L3/L4 iptables rules — IP/CIDR allow/deny

2

Domain Filter

L7 SNI + Host header inspection, wildcards

3

TLS Interception

Per-sandbox CA, decrypt request/response body

4

Guardrails

PII redaction, prompt injection, code security, toxicity, invisible text — redact/block/log

5

Transformation Engine

Regex match/replace rules on body

6

Audit Logger

Metadata + events logged, configurable retention

Early access

Join the waitlist

We're onboarding teams one at a time. Tell us what you're building and we'll get you set up.

This email will be whitelisted for console access.

Support

FAQ