NVIDIA NeMo Guardrails by NVIDIA

# Fix list: NVIDIA NeMo Guardrails

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/nemo-guardrails, the October 2026 research run, assessed 1 October 2026. Grade B, 68.7 out of 100.

This is everything the published grade says the listing lacks, the biggest possible gain to the total first. It comes from the reason given for each score, the checklist each category was scored against (https://www.anchorterminal.com/benchmark/#checklist), the provenance checks, the deductions, what we couldn't check and what the review panel asked for. A fix counts at the next check, once it's public.

For a coding agent working on NVIDIA NeMo Guardrails: work through the items below in the product, its docs and its public pages. Each category gives the reason for its score, with the points each checklist item earned, and the checklist itself, so the gap is the items that earned less than their points. Change the product, not the wording, and keep a note of what you changed and where it's published.

## 1. Security & auth, 62 out of 100, up to 6.7 more on the total

Why it scored 62: Framework reading. No auth of its own, and SECURITY.md says authentication, authorisation, TLS and rate limiting are the deployer's job. Provider keys come from the environment (10 of 30). Tool-input and tool-output rails can block a tool call, but there's no human approval hook, and the LLM-judged tool_safety_check is only on develop since 29 September 2026 (12 of 20). Jailbreak heuristics and model, injection detection and content-safety rails, documented (15). OpenTelemetry tracing to your own backend, opt-in (12 of 15). Disclosure through NVIDIA PSIRT, no bug bounty, no published advisories, and usage telemetry to NVIDIA on by default with a documented opt-out (13 of 20).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-security):

- 0 to 30, the credential model. 30 for OAuth 2.1 with scopes, or scoped and revocable keys with rotation. 20 for plain revocable API keys. 10 for one all-powerful key. 10 off when a secret can travel in a URL query string as a documented option.
- 0 to 20, read-only or least-privilege modes, and confirmation or approval for destructive actions.
- 0 to 15, prompt-injection posture where the tool returns untrusted content (documented mitigations or guidance). A tool that returns no untrusted content gets 10.
- 0 to 15, audit logs or per-call visibility for the operator.
- 0 to 20, a security programme. security.txt or a disclosure policy, a bug bounty, SOC 2 or ISO 27001, advisories handled in public.

Models are read for retention, whether API data trains models (and whether that's off by default), zero-retention options and certifications. Frameworks for telemetry defaults, approval hooks, guardrails and sandboxing.

## 2. Agent ergonomics, 67 out of 100, up to 5.4 more on the total

Why it scored 67: Framework reading. An OpenAI client pointed at the server needs no code, but each rail set needs a config directory with config.yml and prompts (20 of 25). /v1/checks returns a RailOutcome of allow, block or transform since 0.24.0 (15 of 20). Streaming rails fail closed when an action errors, but error responses aren't documented as such (12 of 20). Checks are stateless, but every LLM-based rail is an extra model call to budget and retry (12 of 20). Python only, many defaults, and two Colang versions to choose between (8 of 15).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-ergonomics):

- 0 to 25, context cost. For MCP, the number and size of the tool definitions (25 for ten or fewer compact tools, 15 for 11 to 30, 5 for more than 30, plus up to 10 back for toolsets, dynamic loading or read-only subsets). For APIs, whether responses can be sized (field selection, limits, summaries).
- 20, pagination, filtering and output-size controls.
- 20, actionable, documented error responses, codes and messages an agent can recover from.
- 20, idempotency or safe retries, and for MCP the `readOnlyHint` and `destructiveHint` annotations.
- 15, sensible defaults, few required parameters, and official SDKs in at least two languages.

Models are read for tool use, structured output, prompt caching, context length, batch and SDKs. Frameworks for how much code and how many defaults a tool-calling agent with MCP needs.

## 3. Reliability, 75 out of 100, up to 5 more on the total

Why it scored 75: Local framework reading. Installs from PyPI as nemoguardrails, Python 3.10 to 3.13 stated (20). The README carries passing badges for tests on Linux, Windows and macOS and for lint on the develop branch (25). About 140 open issues and 80 to 100 open pull requests, with bug reports from August 2026 still labelled needs triage beside others already triaged (15 of 25). The changelog follows Keep a Changelog and marks breaking items, six of them in 0.24.0 (15). Version 0.24.1, not 1.0 (0).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-reliability):

Hosted APIs, MCP servers, models and platforms.

- 20, a public status page with component history (Statuspage, Instatus, BetterStack or the vendor's own).
- 0 to 30, the incident record for the last 90 days on that page. 30 for a clean record or trivial incidents only, 20 for minor incidents only, 10 for one major outage (an hour or more of a core API down, or errors across the board), 0 for several. 5 when there's no history we could read, and the note says so.
- 15, rate limits documented with numbers.
- 15, documented 429 or overload handling (Retry-After, backoff guidance), and idempotency keys or safe-retry guidance where writes are involved.
- 10, an SLA published for any paid tier.
- 10, the surface agents use is generally available, not beta or preview.

Local packages, SDKs, frameworks and stdio MCP servers.

- 20, installs from an official package with supported runtimes stated.
- 25, a public CI and test suite, passing on the default branch.
- 0 to 25, open crash or regression issues relative to activity (25 for few and handled, 0 for many, old and unanswered).
- 15, semver discipline and breaking changes called out in a changelog.
- 15, version 1.0 or later, or declared stable.

Protocols are read from their reference implementations, the public facilitators or servers, spec stability and test vectors.

## 4. Schema & documentation, 69 out of 100, up to 5 more on the total

Why it scored 69: Framework reading. Rail configuration is typed in Python and validated on load, and the server speaks the OpenAI chat-completions shape, but we found no published OpenAPI document for /v1/checks (15 of 25). No llms.txt found for the docs (0). The docs describe each rail type and the built-in and third-party rails, and 0.24.0 added IORails, which runs input and output rails without the Colang runtime (15 of 20). YAML configuration with typed models, though Colang 1 and Colang 2 coexist (12 of 15). An examples tree in the repository and per-rail guides, little on error responses (12 of 15). Semver releases and a dated changelog (15).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-schema):

APIs and MCP servers.

- 25, a machine-readable contract (a public OpenAPI file or similar; for MCP, typed JSON Schema inputs on every tool).
- 10, llms.txt or Markdown docs served for agents.
- 0 to 20, descriptions that say what a tool is for, when to use it and when not to, read from the tool definitions in the source or the API reference.
- 0 to 15, typed inputs with enums, constraints and required fields, and no free-form JSON blobs.
- 0 to 15, examples and documented error responses.
- 15, versioning and a public changelog.

Models are read from the API reference, the OpenAPI file, llms.txt, the structured-output and tool-use docs and the model cards. Frameworks from docs a model can follow, typed interfaces, examples and the API reference.

## 5. Payments & pricing, 60 out of 100, up to 5 more on the total

Why it scored 60: Apache-2.0 package you run yourself with nothing to buy from the project, so 20 for pricing, 20 for a free start and 20 for keyless use. No payment protocol (0). The NemoGuard models it can call are priced separately by NVIDIA.

The checklist (https://www.anchorterminal.com/benchmark/#checklist-payments):

The published rubric, also on the [x402 page](https://www.anchorterminal.com/x402/).

- 40, a machine payment protocol (x402, MPP or L402) on the tool's own endpoints. 10 to 30 when it covers only some endpoints or only goes through a third party, and the note says which.
- 20, per-call or per-unit pricing published without a login. 10 for public plan-only pricing, 0 for "contact sales" or prices behind a login.
- 20, a free tier or trial that doesn't need a card.
- 20, autonomous onboarding, meaning an agent can get access without a person signing up in a browser (keyless use, x402, a programmatic key API).

Payment platforms and agent wallets rarely charge for their own API over a machine protocol, so the first line has steps for them, and the highest one that applies counts. 40 when x402, MPP or L402 runs on all their own endpoints, 30 when it runs on part of their own API, 25 when their merchants can accept one, 20 for running a facilitator, 15 for paying as a buyer, and 0 when the only protocol is their own. Merchant acceptance sits above a facilitator because the platform's own customers can charge agents through it, while a facilitator settles for sellers who wire up the protocol themselves. The counter-argument (a facilitator does more for the protocol as a whole) has a point. Each note says which step applied.

Open-source software you run yourself is scored on its hosted or paid option if it has one. A free, self-hosted package with nothing to buy gets 20, 20 and 20 for the last three lines, and 0 to 40 for the first only if it ships a payment protocol.

## 6. Transparency & trust, 71 out of 100, up to 2.5 more on the total

Made of editorial 82, provenance 59.

Why it scored 71: Apache-2.0 with a third-party licence file (30). The telemetry page lists what's sent (version, configuration, which capabilities are switched on, deployment type) and what isn't (prompts, completions, messages, keys, endpoints). We didn't read the code to confirm it (25 of 30). Breaking items are marked per release, but there's no deprecation policy with dates (12 of 20). Telemetry with heartbeats every 10 minutes is on by default, opt-out with NEMO_GUARDRAILS_NO_USAGE_STATS=1, DO_NOT_TRACK=1 or a do_not_track file, and off in CI (15 of 20).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-transparency):

- 0 to 30, source availability and licence clarity. 30 for open source under an OSI licence, 15 for closed with clear terms, 0 for unclear terms.
- 0 to 30, data handling and retention statements that agree with each other (privacy policy, DPA, retention periods, subprocessors).
- 0 to 20, a deprecation policy or notices with dates.
- 0 to 20, telemetry disclosed with an opt-out (local software), or subprocessors and data locations disclosed (hosted).

The other half of Transparency and trust is the provenance score, computed from checked facts (below). The category score is the mean of the two.

Provenance checks not met in full (half of this category, computed from checked facts):

- Domain age: nvidia.com, no registry record we could read (0 of 15)
- Status page: not found (0 of 10)
- security.txt: could not be fetched (0 of 10)

## 7. Maintenance & community, 80 out of 100, up to 1.8 more on the total

Why it scored 80: 0.24.1 on 16 September 2026 (30). Two releases in the last 90 days, 0.24.0 on 25 August and 0.24.1, with 0.23.0 on 1 July just outside (10 of 20, our batch rule for two). Issues get triage labels, but about a dozen opened in August still read needs triage (15 of 25). Current package on PyPI (15). CI on three operating systems and supported runtimes stated (10).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-maintenance):

- 0 to 30, time since the last release, or the last published model or API change for a closed service. 30 within 30 days, 20 within 90, 10 within 180, 0 older.
- 20, at least three releases or dated changelog entries in the last 90 days.
- 0 to 25, responsiveness. Issues and pull requests answered on GitHub (the open issues and how recent the replies are). For closed services, a public changelog and a support or community channel that answers, 0 to 15.
- 15, presence in the official MCP registry under a verified namespace (MCP servers), or current official SDKs (APIs and models).
- 10, package health, current dependencies and CI.

Models are read for deprecation notice periods and model churn rather than release counts.

## What we couldn't check

What we couldn't read counted as absent. Publishing it on a page a plain HTTP fetch can read (not only in a browser) lets the next check count it.

- Which release introduced the default-on telemetry and heartbeat. The telemetry page doesn't say.
- Whether the server publishes an OpenAPI document for /v1/checks.
- Whether the develop-branch tool_safety_check rail ships in the next release.

## Weaknesses

- Usage telemetry and a heartbeat every 10 minutes to NVIDIA by default
- Six breaking changes in 0.24.0, and the project is still pre-1.0
- No authentication on the server, by design
- Every LLM-based rail adds a model call per turn
- About 140 open issues, some from August still untriaged

## What costs an agent a turn today

The notes we give agents before they call it. Each one is a workaround an agent shouldn't need.

- Set NEMO_GUARDRAILS_NO_USAGE_STATS=1 before import unless you want deployment metadata sent to NVIDIA every 10 minutes
- Use IORails for plain input and output checks. LLMRails and Colang are for dialogue flows a tool-calling agent rarely needs
- Pin nemoguardrails==0.24.1. 0.24.0 changed message passing to messages= and removed inline config from /v1/checks
- Call /v1/checks with a config_id loaded on the server and branch on the RailOutcome
- Put the server behind your own gateway. It has no auth or rate limiting

## What the review panel asked for

- Publish an OpenAPI document for /v1/checks
- Document the error responses
- an approval hook for tool rails
- telemetry off by default

## When it's done

Send what changed and where it's published as a dispute (https://www.anchorterminal.com/builders/#disputes, or `POST https://www.anchorterminal.com/api/v1/contact` with `"kind": "dispute"`). Disputes are answered in public, and the listing is checked again by the same checklist. Paying for an audit or a listing claim changes nothing here.