
Argent with LLM Guard
Already run Protect AI's LLM Guard? Point Argent at it, and its scanners make the injection and content-safety call for every prompt, and every reply when reply checks are on, that crosses the gateway, from any client and any model, under one Argent policy.
What Argent does with LLM Guard
Your scanners make the call. Argent enforces it at the gateway and records it.
LLM Guard is Protect AI's open-source security toolkit for LLM interactions, and it can be run as an API (LLM Guard API docs). Argent connects to that API as a guardrail.
- Prompts. Each prompt is sent to LLM Guard's prompt scanners before it reaches the model.
- Replies too. With the egress lane on, replies are checked as well. A classifier flags, it cannot cut out words, so under the default reply setting a flagged reply goes through unchanged with a note on the signed record; under block, the whole reply is replaced with a safe error.
- LLM Guard's verdict is the block. When LLM Guard says the text is not valid, Argent treats that as a hard block.
- Scores for the record. Each scanner's score is mapped onto Argent's own category names, so verdicts read the same as every other engine in the console and on the signed record.
How its scanners map
Each LLM Guard scanner Argent recognizes feeds one or more Argent categories.
| LLM Guard scanner | Argent category |
|---|---|
| PromptInjection | prompt_injection and jailbreak |
| Toxicity | toxicity |
| Secrets | secrets |
| Anonymize | pii |
| BanTopics | off_topic |
| Code, BanCode | code |
| Gibberish | gibberish |
A scanner with no Argent category is left out of coverage. If two scanners feed the same category, the higher score counts.
How it connects
Argent calls your LLM Guard API. LLM Guard never calls Argent.
| Direction | Argent's gateway calls your LLM Guard API |
| Endpoints | /analyze/prompt for prompts, /analyze/output for replies, /healthz for the reachability check, all under the base URL you give |
| Sign-in | Optional. If your LLM Guard API requires a bearer token, Argent sends one kept in your vault |
| TLS | Verified by default. For a self-signed server you can pin your own CA certificate, and verification stays on |
| Time limit | 1.5 seconds per check by default, adjustable up to 30 seconds |
| Scanner choice | An optional list of scanners to skip on each call, passed to LLM Guard as given |
| Network | Every call goes through Argent's guarded outbound client |
What you configure
Connect it once, then point the tenant's classifier policy at it.
- Run the LLM Guard API where the gateway can reach it, with the scanners you want.
- Store the token, if any, in your vault: HashiCorp Vault or OpenBao, AWS Secrets Manager, CyberArk Conjur or Infisical. Argent keeps only the reference.
- Add the connector. On the Integrations page, under Guardrails, choose LLM Guard and enter the base URL, plus any time limit, CA certificate, scanners to skip and token reference. Then Connect and Test.
- Use it in the policy. In the tenant's classifier policy, choose your guardrail as the scoring backend, pick this connector and enable the categories you want enforced.
- Watch first. In observe mode, verdicts are scored and recorded without blocking. Switch to block when the record looks right.
When LLM Guard cannot answer
A guardrail that cannot answer is never read as a pass.
- Could not measure is not clean. If the guardrail times out, cannot be reached, rejects Argent's credential or sends back an answer Argent cannot read, the check counts as not measured. What happens next is the policy's on-timeout setting: deny, or allow with a note on the signed record.
- No silent fallback. A tenant that hands the decision to its own guardrail is resolved to that guardrail only. Argent never quietly scores it with a different engine instead.
- Every enabled category must be covered. If the policy turns on a category the vendor did not evaluate, that category fails closed. It is never scored as zero.
- One policy language. Policies name categories from Argent's closed set of eleven: prompt_injection, jailbreak, toxicity, hate, sexual, violence, pii, secrets, off_topic, gibberish and code. A policy write that names anything else is refused.
- Credentials stay references. A token or key lives in your vault. Argent stores only the reference and resolves it when it connects, and a reference it cannot resolve stops the connection.
- Connectors are an administrator's job. The Integrations page is gated by console role, like every console route.
Where this fits
Pages that go deeper.
- Other engines: Amazon Bedrock Guardrails as your guardrail, or the ONNX injection classifier inside your deployment.
- Capabilities: Prompt Injection Defense, Response Screening, Watch Before You Enforce, Session Replay and all capabilities.
- Goals: threat defense, data loss prevention, sovereign AI and all goals.
- Use cases: governing customer-facing AI assistants, rolling out copilots and chat safely, securing AI coding agents, controlling agent tool access and AI in regulated and air-gapped environments.
- Where it runs: air-gapped on-premises, private cloud or your public cloud VPC. Compare them on deployment options.
- Industries: Financial Services, Healthcare, Insurance, Public Sector, Retail and all industries.
- More integrations: every system Argent connects to, grouped by type, on the integrations page.
Want this running on your own use cases?
All downloads