Senthex — AI Firewall for LLM APIs

Senthex sits between your application and the LLM API. Every request passes through 24 shield modules: heuristic injection detection, Presidio PII scanning, intent classification, bypass detection, data classification, budget enforcement, and more. Responses are scanned for secret leaks, canary exfiltration, toxicity, and dangerous output patterns.

SSE streaming is fully transparent for both OpenAI (data: {...}) and Anthropic (event: ...\ndata: {...}) formats. Shield analysis runs inline, chunk-by-chunk. Logging is fire-and-forget via asyncio.create_task.

The first LLM firewall designed for autonomous agents, not just human developers. Agents can read, react, and configure — programmatically, without human intervention.

from senthex import SenthexClient

client = SenthexClient(senthex_key="sx_...")

# Agent monitors its own security posture
usage = client.usage()
if usage.budget_remaining_eur < 1.0:
    agent.reduce_activity()

# React to shield blocks — machine-readable
resp = client.chat(messages=messages)
if resp.shield_status == "blocked":
    agent.handle_block(resp.block_reason)

# Canary integrity — know if your prompt leaked
if resp.canary_triggered:
    agent.alert_and_rotate_prompt()

# Check trust level — adapt if flagged as bypass
if resp.trust_level == "reduced":
    agent.reset_session()

Traditional firewalls have fixed thresholds. Attackers find the edge and stay just below. Senthex moves the edge. Every suspicious request makes the next one harder to pass.

Every detection is heuristic-based. No LLM in the loop — too slow, too recursive. Pattern matching, scoring, and NER. Configurable thresholds and actions per project.

Not all risks are solvable at the proxy layer. Here's an honest breakdown of what Senthex covers, what's partial, and what requires application-level controls.

Change one line. All shields apply automatically.

from openai import OpenAI

client = OpenAI(
    api_key="sk-...",
    base_url="https://app.senthex.com/v1",
    default_headers={"X-Senthex-Key": "snx-..."}
)

response = client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": "Hello"}]
)

import anthropic

client = anthropic.Anthropic(
    api_key="sk-ant-...",
    base_url="https://app.senthex.com",
    default_headers={"X-Senthex-Key": "snx-..."}
)

message = client.messages.create(
    model="claude-opus-4-6",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Hello"}]
)

from openai import OpenAI

client = OpenAI(
    api_key="...",
    base_url="https://app.senthex.com/v1",
    default_headers={
        "X-Senthex-Key": "snx-...",
        "X-Senthex-Provider": "mistral"
    }
)

response = client.chat.completions.create(
    model="mistral-large-latest",
    messages=[{"role": "user", "content": "Hello"}]
)

from openai import OpenAI

client = OpenAI(
    api_key="AIza...",
    base_url="https://app.senthex.com/v1",
    default_headers={
        "X-Senthex-Key": "snx-...",
        "X-Senthex-Provider": "google"
    }
)

response = client.chat.completions.create(
    model="gemini-2.0-flash",
    messages=[{"role": "user", "content": "Hello"}]
)

from openai import OpenAI

client = OpenAI(
    api_key="sk-or-...",
    base_url="https://app.senthex.com/v1",
    default_headers={
        "X-Senthex-Key": "snx-...",
        "X-Senthex-Provider": "openrouter"
    }
)

response = client.chat.completions.create(
    model="anthropic/claude-opus-4-6",
    messages=[{"role": "user", "content": "Hello"}]
)

curl https://app.senthex.com/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer sk-..." \
  -H "X-Senthex-Key: snx-..." \
  -d '{
    "model": "gpt-4o",
    "messages": [{"role": "user", "content": "Hello"}]
  }'

Every request is logged with metadata only by default — timestamps, scores, PII types detected, trust level, latency. Full request/response bodies are never stored.

Currently in free beta. Request access below — no credit card required.

Senthex is in free beta. Looking for developers and teams who use LLM APIs in production to test it and provide feedback. What works, what's missing, what would make this worth paying for.

Now with 24 shields: multi-turn tracking, bypass detection with trust levels, intent classification, toxicity scoring, data classification, automatic prompt hardening, file upload scanning, and more.