                        
                        AI AGENT TOOL FIREWALL • VERIFIABLE SHA-256 AUDIT LOGS
                    

Keep full control
of your AI agents
in production

Boundry intercepts every tool call, enforces your permissions, detects prompt injections, and seals every action in a verifiable audit log.

Start for free Explore the live interactive demo console →

Supervision et contrôle des appels d'outils des agents IA avec Boundry

Intercept every
tool call before
execution

Autonomous AI agents interact directly with your production databases and internal APIs. Boundry enforces tool whitelists locally in the SDK (< 1.4 ms at p95), detects known prompt injections, and records every action in a signed SHA-256 chain.

Designed for EU AI Act (Art. 12 & 14) GDPR Aligned SOC 2: in preparation ISO 27001: roadmap

Fast integration with the Boundry Python SDK — configurable per agent in observe-only or active-blocking mode.

console.boundry.cloud

[UTC 17:58:12] [LOCAL SDK] Whitelist check passed (0.7ms): tool='read_faq_database'
[UTC 17:58:12] [ENGINE] Evaluating prompt context against OWASP LLM01 rules (21ms)...
[UTC 17:58:13] [PASS] Signed SHA-256 entry appended. Tamper-evident chain verified.
[UTC 17:58:15] [WARN] Tool 'send_email' blocked locally (not in agent whitelist).
[UTC 17:58:16] [QUARANTINE] Suspected indirect prompt injection. Action held for review.

Supervise, control, and audit your AI agents

A defense-in-depth architecture combining prompt injection detection, deterministic tool permissions, and verifiable audit evidence.

Defense-in-Depth Detection

Detects and blocks known prompt injections (OWASP LLM01 rules and heuristics), and limits the blast radius of novel attacks through tool whitelisting and quarantine. Start in observe-only mode, then enable blocking when ready.

Tool Whitelist & Quarantine

Every agent has a strict tool whitelist evaluated locally in the Python SDK. Any out-of-policy call is stopped or quarantined, with per-agent fail-mode configuration (on_unreachable).

Tamper-Evident Audit Logs

Tamper-evident logging via per-organization signed and anchored SHA-256 hash chains: any modification or deletion is detectable offline with our open-source CLI verifier, supporting your EU AI Act and GDPR audits.

Live Tool Permission Control (Local SDK Whitelist)

Test the deterministic tool permission switch enforced prior to execution:

get_weather_data

send_invoice_email

Tamper-Evident SHA-256 Hash Chain Simulator

// BLOCK N (Signed log verifiable via boundry_verify.py)
action: read_db {query: 'SELECT email FROM users'}
prev_hash: e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
hash SHA-256: -

Launch Partners Program

Deploying AI agents in production?

We are selecting 5 partner engineering teams: get 3 months of the Pro plan for free and direct engineering support with the founders in exchange for candid technical feedback.

Apply to the program →

Talk directly with the engineers

Questions about the architecture, on_unreachable fail-modes, or the Launch Partners program? The Neko Labs engineering team replies within 48 business hours.

Keep full control of your AI agents in production

Intercept every tool call before execution