Claude Code

Use Claude Code with SecureAI — full SMLTP, DLP, and PII enforcement for AI-assisted development

Claude Code — IDE Integration

Claude Code is Anthropic’s agentic coding assistant. By pointing it at SecureAI’s Anthropic-compatible proxy, every prompt your developers send passes through SMLTP, DLP, PII scanning, Prompt Shield, rate limiting, and model governance — without any changes to the developer’s workflow.

---

How it works

SecureAI exposes an Anthropic Messages API-compatible endpoint at:

https://<your-host>/api/claude-code

Claude Code natively supports redirecting all traffic to a custom base URL via the ANTHROPIC_BASE_URL environment variable. When configured, it speaks its native Anthropic protocol directly to SecureAI — no local proxy or wrapper script required.

Developer machine
    │
    │  Anthropic Messages API (POST /v1/messages)
    ▼
SecureAI  (/api/claude-code)
    │
    ├── API key auth
    ├── SMLTP bundle + policy hash
    ├── DLP scan (input)
    ├── PII scan + graduated enforcement
    ├── Prompt Shield (injection detection)
    ├── Model governance & rate limiting
    │
    ▼
OpenRouter / Self-Hosted LLM

All activity appears in the SecureAI admin dashboard under AI Gateway → Audit Logs, attributed to the developer’s API key.

---

Step 1 — Install Claude Code

macOS / Linux / WSL

curl -fsSL https://claude.ai/install.sh | bash

Windows PowerShell

irm https://claude.ai/install.ps1 | iex

---

Step 2 — Generate an API key

1. Go to Admin → API Keys (https://<your-host>/admin/apis)
2. Click Create API Key
3. Set a name (e.g. claude-code-dev-jane)
4. Configure:
   • Allowed models — select which LLMs the key can use (see available models)
   • SMLTP policy — select the compliance policy (internal, confidential, hipaa, etc.)
   • Rate limits as required
5. Copy the sk-… value — it is shown only once

---

Step 3 — Configure Claude Code

The recommended approach is to add the settings to the project-level local settings file (.claude/settings.local.json), which is git-ignored by default.

.claude/settings.local.json

{
  "env": {
    "ANTHROPIC_BASE_URL":  "https://<your-host>/api/claude-code",
    "ANTHROPIC_AUTH_TOKEN": "sk-<your-secureai-api-key>",
    "ANTHROPIC_API_KEY":   "",

    "ANTHROPIC_DEFAULT_OPUS_MODEL":   "anthropic/claude-opus-4.6",
    "ANTHROPIC_DEFAULT_SONNET_MODEL": "anthropic/claude-sonnet-4.6",
    "ANTHROPIC_DEFAULT_HAIKU_MODEL":  "google/gemini-3-flash-preview",
    "CLAUDE_CODE_SUBAGENT_MODEL":     "anthropic/claude-sonnet-4.6",

    "CLAUDE_CODE_SKIP_FAST_MODE_ORG_CHECK": "1"
  }
}

ANTHROPIC_API_KEY must be explicitly set to an empty string to prevent Claude Code from trying to authenticate directly with Anthropic.

Shell profile (alternative)

You can also add the variables to ~/.zshrc, ~/.bashrc, or your PowerShell $PROFILE instead of the settings file. The settings file is preferred for team projects so every developer inherits the same configuration automatically.

---

Step 4 — Verify the connection

Start Claude Code from your project directory:

cd /path/to/your/project
claude

Run /status inside the session. You should see:

Auth token:          ANTHROPIC_AUTH_TOKEN
Anthropic base URL:  https://<your-host>/api/claude-code

That confirms all traffic is routing through SecureAI.

---

Available models

SecureAI exposes the same model catalog available in the chat interface. Use any id value from the table below in the env vars above.

Anthropic (Claude)

Model ID	Display name
anthropic/claude-sonnet-4.6	Claude Sonnet 4.6
anthropic/claude-opus-4.6	Claude Opus 4.6

OpenAI

Model ID	Display name
openai/gpt-5-mini	GPT-5 Mini
openai/gpt-5.1-codex	GPT-5.1-Codex
openai/gpt-5.1	GPT-5.1
openai/gpt-5.2	GPT-5.2
openai/gpt-5.3-codex	GPT-5.3-Codex

Google (Gemini)

Model ID	Display name
google/gemini-3-flash-preview	Gemini 3 Flash Preview
google/gemini-3.1-pro-preview	Gemini 3.1 Pro Preview

Meta (Llama)

Model ID	Display name
meta-llama/llama-4-maverick	Llama 4 Maverick
meta-llama/llama-4-scout	Llama 4 Scout
meta-llama/llama-3.3-70b-instruct	Llama 3.3 70B Instruct

Mistral

Model ID	Display name
mistralai/mistral-large-2512	Mistral Large 3 2512
mistralai/ministral-14b-2512	Ministral 14B 2512
mistralai/mistral-nemo	Mistral Nemo
mistralai/mistral-7b-instruct	Mistral 7B Instruct

DeepSeek

Model ID	Display name
deepseek/deepseek-r1-0528	DeepSeek R1 0528
deepseek/deepseek-v3.2	DeepSeek V3.2

xAI (Grok)

Model ID	Display name
x-ai/grok-4	Grok 4
x-ai/grok-code-fast-1	Grok Code Fast 1
x-ai/grok-4.1-fast	Grok 4.1 Fast

Qwen

Model ID	Display name
qwen/qwen3-coder	Qwen3-Coder
qwen/qwen3-coder-next	Qwen3 Coder Next
qwen/qwen3-235b-a22b-2507	Qwen3 235B A22B 2507
qwen/qwen3.5-397b-a17b	Qwen3.5 397B A17B

Self-Hosted (Remote SMLTP Endpoints)

Self-hosted models registered as active remote SMLTP endpoints are automatically available. Their IDs follow the pattern self-hosted/<model-name>. Run /status in Claude Code or call GET /api/claude-code/v1/models to see the live list.

---

Claude Code model roles

Claude Code uses different models for different internal tasks. Map each role to any model from the catalog above:

Environment variable	Role	Recommended
ANTHROPIC_DEFAULT_OPUS_MODEL	Complex reasoning, main agentic loop	anthropic/claude-opus-4.6
ANTHROPIC_DEFAULT_SONNET_MODEL	General coding tasks	anthropic/claude-sonnet-4.6
ANTHROPIC_DEFAULT_HAIKU_MODEL	Quick completions, tool calls	google/gemini-3-flash-preview
CLAUDE_CODE_SUBAGENT_MODEL	Spawned sub-agent tasks	anthropic/claude-sonnet-4.6

You can mix providers freely. For example, use Gemini for fast/cheap sub-tasks and Claude for the main reasoning loop:

"ANTHROPIC_DEFAULT_OPUS_MODEL":   "anthropic/claude-opus-4.6",
"ANTHROPIC_DEFAULT_SONNET_MODEL": "x-ai/grok-code-fast-1",
"ANTHROPIC_DEFAULT_HAIKU_MODEL":  "google/gemini-3-flash-preview",
"CLAUDE_CODE_SUBAGENT_MODEL":     "deepseek/deepseek-v3.2"

---

Security enforcement

Every request through the proxy is subject to the full SecureAI security stack:

Layer	What it does
SMLTP	Attaches bundle ID, policy hash, trace ID, and egress enforcement to every call
DLP	Scans prompts for data-loss patterns (secrets, credentials, confidential docs)
PII	Detects and redacts personally identifiable information; graduated enforcement tiers
Prompt Shield	Detects prompt injection attempts
Model governance	Enforces allowed models, provider allowlists, and residency rules
Rate limiting	Per-key and per-user limits enforced via Redis
Audit logs	Full per-request audit trail in AI Gateway → Audit Logs

The SMLTP policy is inherited from the API key configuration. To change the policy for a key, go to Admin → API Keys, edit the key, and select a different SMLTP Policy.

---

Troubleshooting

/status still shows api.anthropic.com

Claude Code has cached credentials from a previous login. Run /logout inside a Claude Code session, then restart:

claude
❯ /logout

401 Unauthorized

• Verify the sk-… key is active in Admin → API Keys
• Check that ANTHROPIC_API_KEY is set to an empty string ("")

Model not available

• Check the API key’s Allowed Models list in Admin → API Keys
• Call GET /api/claude-code/v1/models (with your Bearer sk-… token) to see exactly what the key can access

DLP / PII block

If a prompt is blocked, Claude Code will receive an error response with a message explaining the violation. The incident is logged in Admin → Incidents and visible in AI Gateway → Audit Logs.