Implementation Workflows

Two tailored workflows for complete AI security.

SecureMID supports both **Portal Web UI & REST API-based testing** for interactive exploration and scheduled pentesting, and **Headless CLI-based gating** for automated CI/CD pipelines.

Track A: Portal UI & REST API Testing Track B: CI/CD CLI Gating (securemid-cli)

TRACK A // CUSTOMER PORTAL WEB UI & REST API TESTING

Interactive testing via Customer Portal & REST APIs

Designed for security teams, AI engineers, and compliance leads who configure, run, and inspect evaluations visually in the web portal or orchestrate them programmatically via our HTTP API.

STEP 01 // PORTAL REGISTRATION & ASSET ONBOARDING

Connect Your Agent Endpoint via Web UI or REST API

In the Customer Portal **Asset Registry**, register your AI agent endpoint URL (HTTP REST or WebSocket) and specify staging/production tags. Alternatively, programmatically register assets using our REST API:

REST API // POST /api/v1/agents

1	POST /api/v1/agents
2	Host: app.securemid.com
3	Authorization: Bearer <SECUREMID_API_KEY>
4	Content-Type: application/json
5
6	{
7	"name": "FinBot-Production",
8	"endpoint_url": "https://agent-staging.internal:9000/v1/chat",
9	"environment": "staging",
10	"auth_type": "bearer",
11	"model_identifier": "gpt-4o-mini-financial",
12	"protocol": "http_rest"
13	}

STEP 02 // THREAT BATTERY CONFIGURATION

Select Attack Batteries & Safety Invariants in the Dashboard

Use the Portal UI or API to select threat matrices (OWASP LLM01 Prompt Injection, LLM05 Tool Abuse, SSRF, RAG Poisoning) and configure conversation depth:

Standard Fast ScanCore prompt injection & PII leakage checks. 2–5 minutes completion.

Deep Crescendo ScanMulti-turn adversarial probing of tool parameter authorization limits.

Custom Safety InvariantsEnforce proprietary organizational compliance rules and output syntax.

STEP 03 // AIR-GAPPED EVALUATION EXECUTION

Trigger Distributed Scan via Portal API or 1-Click Launch

Celery workers simulate attacks asynchronously while our local fine-tuned judge evaluates responses in **4.20 ms**. Zero prompt data leaves your perimeter:

PORTAL REST API // POST /api/v1/assessments/start

1	POST /api/v1/assessments/start
2	Host: app.securemid.com
3	Authorization: Bearer <SECUREMID_API_KEY>
4	Content-Type: application/json
5
6	{
7	"agent_id": "agt_88192a7b",
8	"batteries": [
9	"owasp_llm01_prompt_injection",
10	"owasp_llm05_tool_abuse",
11	"owasp_llm08_ssrf_metadata",
12	"crescendo_multi_turn_escalation"
13	],
14	"max_conversation_depth": 10,
15	"fail_on_severity": "HIGH",
16	"judge_mode": "air_gapped_local"
17	}

Air-Gapped Local Guarantee: `SECUREMID_ALLOW_CLOUD_FALLBACK=false` ensures zero prompt leakage.

STEP 04 // FINDINGS INSPECTOR & 1-CLICK RETEST

Review Captured Payloads & Retest in Customer Portal

Inspect line-by-line developer remediation code and cryptographic SHA-256 evidence. Once patched, click **"Run 1-Click Retest"** in the portal to confirm vulnerability closure.

Schedule Remediation Pilot

TRACK B // AUTOMATED CI/CD PIPELINE GATING (CLI-BASED)

Headless build gating with `securemid-cli`

Designed for DevOps, platform engineers, and CI/CD pipelines. Automatically block vulnerable agent prompt merges and tool schema changes before production release.

CLI SETUP // PYTHON PACKAGE

1. Install the CLI in your CI runner or pre-commit hook

TERMINAL // PIP INSTALL

python -m pip install securemid-cli

2. Execute headless adversarial scan with severity exit codes

The CLI exits with code 1 when a vulnerability matches or exceeds the specified severity threshold, immediately failing the CI/CD build.

TERMINAL // RUN SCAN

securemid-cli scan \
  --agent-name "FinBot-Agent" \
  --endpoint-url "http://localhost:9000/v1/chat" \
  --org-id "${{ secrets.SECUREMID_ORG_ID }}" \
  --fail-on-severity HIGH \
  --wait

3. GitHub Actions Pipeline Integration

GITHUB ACTIONS // .github/workflows/ai-security.yml

1	# .github/workflows/ai-security.yml
2	name: AI Security Gate
3	on: [pull_request]
4
5	jobs:
6	securemid-eval:
7	runs-on: ubuntu-latest
8	steps:
9	- uses: actions/checkout@v4
10	- name: Set up Python
11	uses: actions/setup-python@v5
12	with:
13	python-version: '3.11'
14
15	- name: Install SecureMID CLI
16	run: python -m pip install securemid-cli
17
18	- name: Run Headless Adversarial Evaluation Gate
19	env:
20	SECUREMID_API_KEY: ${{ secrets.SECUREMID_API_KEY }}
21	run: \|
22	securemid-cli scan \
23	--agent-name "FinBot-Agent" \
24	--fail-on-severity HIGH \
25	--output-report ./security-report.json \
26	--wait

Operational Boundaries & Troubleshooting

Context Window Boundaries

The Security Judge model evaluates up to 2,048 tokens per assertion. For extensive agent trajectories exceeding 10 turns, intermediate tool traces are summarized to prevent truncation.

Air-Gapped Daemon Fallback

If the local judge daemon becomes unreachable, SecureMID safely records an UNKNOWN status rather than silently egressing sensitive prompts to public cloud providers.

Start Your Security Assessment

Make AI security part of your workflow.

Start exploring how SecureMID can help your team autonomously assess, red-team, and evaluate AI systems before they reach production.

Request Pilot Demo Explore Platform

Free Community Tier Available•Zero Cloud Token Egress•Turnkey Docker & CLI Gating