AI GATEWAY · MCP TOOLS · SECURITY · AUDIT
One layer between your apps and every AI agent.
Route LLM calls to any provider. Execute MCP tools with per-key access control. Scan every request, response, and tool output for injection, PII, and off-topic drift. Sign and audit every step.
Same code. Gained everything.
Change two lines. Gain routing, tools, policy, and audit.
Keep your SDK. Keep your models. Keep your code. Verosek is a drop-in replacement for the OpenAI, Anthropic, and Gemini SDKs.
import openai
client = openai.OpenAI(
api_key="vsk_...", # Verosek virtual key
base_url="http://your-gateway/v1", # ← only line that changed
)
response = client.chat.completions.create(
model="gpt-4o",
messages=[{"role": "user", "content": "How many users are in the database?"}],
)
# Tools handled internally. Policy enforced. Every step audited.
print(response.choices[0].message.content)Architecture at a glance.
One request. Every layer. Under 30 ms of overhead.
Every call flows through virtual-key identity, Shield pre-scan, cross-SDK translation, the provider, the policy engine, tool execution, post-scan, and a signed audit entry.
Four layers. One product.
Routing, tools, security, and audit — without the integration gap.
Layer 01
AI Gateway
Route any SDK to any provider. Twelve OpenAI endpoints plus native Anthropic and Gemini. Cross-SDK translation on the fly.
Learn moreLayer 02
MCP Tools
Fifty tool connectors across databases, SDLC, knowledge, search, automation, business, and observability. DBeaver-style access control for every one.
Learn moreLayer 03
Verosek Shield
24 security checks across input, output, tool output, and session drift. Deterministic where possible. Offline ML everywhere.
Learn moreLayer 04
Audit Vault
Tamper-evident HMAC-SHA256 trace. Decision receipts that verify offline with one CLI command.
Learn moreWhat Verosek catches.
The four attack vectors nobody else structurally handles.
Prompt injection + jailbreak
User prompts contain hidden instructions that override the system prompt.
Multilingual classifier catches direct injection and jailbreak in one forward pass.
Indirect injection via tool output
A SQL row or Slack message contains instructions that the model reads as if you wrote them.
Post-tool scan runs the same classifier against every tool output before it reaches the model.
PII and secret exfiltration
The model leaks customer data in its reply, or a tool reads it into the session.
A multilingual PII engine plus 17 secret-regex patterns scan inputs, outputs, tool results, and cumulative session bytes.
Off-topic drift + hallucination
Agent answers about topics the key never authorised — or fabricates grounded-looking output.
Per-key topic centroids plus an offline grounding verdict flag drift and ungrounded claims.
Receipts, not claims.
Every decision is explainable. Every trace is verifiable offline.
An auditor verifies each receipt with one CLI command — no network call back to Verosek.
{
"trace_id": "trace_01HQ5V8K3P7Z2MJR9",
"decision": "BLOCK",
"key_id": "vsk_prod_04",
"tool": "postgres.query",
"score": 8,
"triggered_checks": [
{
"id": "CHK-002",
"detail": "DELETE not in allowed_operations [SELECT]"
}
],
"signed_at": "2026-04-15T09:41:22Z",
"signature": "hmac-sha256:a7b9c2…4f8e"
}{
"decision": "BLOCK",
"score": 8,
"triggered_checks": [
{
"id": "CHK-002",
"detail": "DELETE not in allowed_operations [SELECT]",
"score_contribution": 8
}
],
"modifications_applied": []
}$ verosek-verify-receipt receipt.jsonreading receipt.json ... signature: hmac-sha256:a7b9c2...4f8e signed at: 2026-04-15T09:41:22Z request_hash match: ok triggered_checks well-formed: ok signature verifies: ok VALID
Three tiers. No surprises.
Start on a 15-day free trial. Graduate when your compliance team asks for evidence.
Trial
No card required. 2 virtual keys, 2 MCP connections, baseline Shield.
- 2 virtual keys
- 50k requests, 50k Shield scans
- Baseline Shield profile
- Community support
Team
Managed cloud. Predictable overage. Baseline + strict Shield profiles.
- 25 virtual keys, 5 MCP connections
- 500k requests / 500k scans / month
- 5 users with admin / member / auditor roles
- Email support, 24-hour response
Enterprise
Annual. Unlimited everything. Managed cloud, self-hosted, or air-gapped.
- Unlimited keys, connections, users
- Custom RBAC, 90+ day audit retention
- HMAC-signed receipts, custom PII recognizers
- Named TAM, 4-hour SLA
Put a governance layer in front of your agents.
FAIL_CLOSED by default. Every tool call scanned. Every decision signed.