# LlamaFirewall vs ShieldGemma

> ShieldGemma scores 54 (C) to LlamaFirewall's 50.1 (D) for policy enforcement. Prices, MCP, x402, uptime and agent notes side by side.

- Canonical: https://www.anchorterminal.com/compare/llamafirewall-vs-shieldgemma
- Markdown: https://www.anchorterminal.com/compare/llamafirewall-vs-shieldgemma.md (~4,850 tokens)
- Site index for agents: https://www.anchorterminal.com/llms.txt (the key docs and every listing in one file, at most 16 MB, about 4 million tokens: https://www.anchorterminal.com/llms-full.txt)
- API: https://www.anchorterminal.com/api/v1/index.json
- MCP: https://www.anchorterminal.com/mcp (the whole catalogue as tools, no key)
- DNS: dig +short TXT <slug>.grade.anchorterminal.com (the grade as a DNS TXT record, for a sandbox with no HTTP where names still resolve; also top., best.<category>. and <a>.vs.<b>. under the zone, and the same under grade.letme.dev)
- Licence: CC BY 4.0, credit Anchor Terminal and link this page (https://www.anchorterminal.com/terms/#data)
- Updated: 2026-11-07

ShieldGemma scores 54 (C) on agent readiness against LlamaFirewall's 50.1 (D), and leads in 4 of 7 scored categories. LlamaFirewall leads on transparency & trust. Both do policy enforcement.

- LlamaFirewall: grade D, 50.1/100, rank #1077 of 1510. Markdown https://www.anchorterminal.com/tools/llamafirewall.md · JSON https://www.anchorterminal.com/api/v1/tools/llamafirewall.json
- ShieldGemma: grade C, 54/100, rank #927 of 1510. Markdown https://www.anchorterminal.com/tools/shieldgemma.md · JSON https://www.anchorterminal.com/api/v1/tools/shieldgemma.json
- Best guardrails and safety filters for AI agents: https://www.anchorterminal.com/best/guardrails/index.md
- All 155 guardrails comparisons: https://www.anchorterminal.com/compare/guardrails/index.md

## Which one, for what

### LlamaFirewall (D)

Good for: A Python agent team that wants injection, hidden-character and generated-code checks in process, is willing to pin dependencies or install from main, and can get the gated weights.

Ahead on:
- Transparency & trust, 62 against 54

Also in its favour:
- Open source

Watch for: No release found after 1.0.3 of May 2025, and no changelog, tags or deprecation notes were found

### ShieldGemma (C)

Good for: Policy-specific content moderation of English text input or output, or of images, on the owner's hardware, with a probability to threshold.

Ahead on:
- Schema & documentation, 56 against 49
- Agent ergonomics, 72 against 60
- Maintenance & community, 40 against 15

Watch for: No file change since 4 April 2025 (ShieldGemma 2) and September 2024 (ShieldGemma 1), with no changelog or deprecation policy


## Why the edge

### Maintenance & community: ShieldGemma +25

- LlamaFirewall (15): The newest version is 1.0.3, set in `pyproject.toml` on 28 May 2025 per a clone of main on 4 November, which holds no tags. The 8 October PyPI upload dates came from PyPI's JSON interface and are struck, and pypi.org did not resolve on 4 November, so no newer release was found (0 of 30). No release in the 90 days from 6 August 2026 (0 of 20). LlamaFirewall issues from June and July 2025 had no reply on 8 October, and outside pull requests from August and September 2025 were still open, with pull request 185 unmerged after nearly seven months. Not re-read on 4 November. Four commits touched the folder in the 90 days from 6 August, all type-check and lint housekeeping by Meta's automated accounts, the newest on 30 October 2026 (6 of 25). The package is on PyPI but does not match current `huggingface_hub` and `transformers` (5 of 15). Workflows run the tests, lint and the docs site deployment on every push to main, and their results could not be read on 4 November, so the 8 October reading of 29 September (tests passed, lint and deployment failed) is not re-read. Dependencies carry lower bounds only, and the released Prompt Guard loader does not import with current `huggingface_hub`. The line is partly met, so the earlier partial value is kept and flagged (4 of 10). 0+0+6+5+4 = 15.
- ShieldGemma (40): Read as a model, on the lines used for Granite Guardian (recency 30, deprecation 12, churn 8, responsiveness 25, packages 15, health 10). The newest file change is ShieldGemma 2 on 4 April 2025, 19 months before the check, and ShieldGemma 1 last changed in September 2024 (0 of 30). No deprecation policy. Both generations stay downloadable, and the ai.google.dev page was marked deprecated with a pointer to gemma.ai without dates for the models (3 of 12). Two generations in 2024 and 2025, each in its own repository, none replaced in place (6 of 8). Google accounts (members of Google's Gemma organisations on the Hub) answered every thread we sampled on the 2B, 9B and ShieldGemma 2 pages, from August 2024 to 22 March 2026, but leave most threads open (16 of 25). Transformers, Keras and Ollama's library load it (12 of 15). No CI, and Ollama's library build was last updated 2 years ago (3 of 10). 0 + 3 + 6 + 16 + 12 + 3 = 40.

### Agent ergonomics: ShieldGemma +12

- LlamaFirewall (60): Read as a classifier an agent calls in process. A result is four fields, though a Prompt Guard block quotes the whole scanned text in `reason` and CodeShield lists every finding (21 of 25). Scanners are chosen per role and two preset use cases exist. Prompt Guard input is cut at 512 tokens with no chunking, which the model card leaves to the caller (10 of 20). A `status` field and a clear message for a missing key exist. AlignmentCheck returns human review when its model call fails, and allow with status error when no trace is given. `scan()` wraps `asyncio.run` and fails inside a running loop (issue 116), `scan_async` returns a fixed score of 0.0 on allow, and none of this is documented (8 of 20). Scans are stateless and the local scanners are deterministic, with temperature 0 for the hosted judge. Each scan builds a new scanner, which reloads the model from disk (14 of 20). Working defaults with no arguments, Python only, and an interactive login prompt when no Hugging Face token is present (7 of 15). 21+10+8+14+7 = 60.
- ShieldGemma (72): Read as a classifier an agent calls. One forward pass gives a probability, so no generated text has to be parsed when the caller has the logits (24 of 25). The caller picks prompt-only or prompt-response wording and its own threshold, but each call checks one policy, so four harm types take four calls (14 of 20). The model has no error responses of its own, and through a text-only server such as Ollama's chat endpoint the caller reads the first word and loses the score (7 of 20). Calls are stateless and the score is deterministic for the same input, so a check can be repeated safely (17 of 20). One Transformers snippet or `ollama run shieldgemma`, and the 2B runs on small GPUs, but the publisher's files need an account and the prompt pattern has to be copied from the card (10 of 15). 24 + 14 + 7 + 17 + 10 = 72.

### Transparency & trust: LlamaFirewall +8

- LlamaFirewall (62): The library is MIT. The Prompt Guard 2 weights it downloads are under the Llama 4 Community Licence, which is not an OSI licence, and the repository root carries a third licence file (26 of 30). Four scanners keep text on the owner's machine. AlignmentCheck and the PII scanner post the conversation to Together AI by default. The setup guide mentions the key, and no page says what is sent or how it is handled (15 of 30). No deprecation policy, changelog or dated notices found (0 of 20). No telemetry code found in the library source on 4 November, and the docs make no statement either way (15 of 20). 26+15+0+15 = 56.
- ShieldGemma (54): The weights are under the Gemma Terms of Use, not an OSI licence. They carry distribution conditions, the Prohibited Use Policy and a clause that lets Google restrict usage remotely where it believes the terms are breached. The terms are dated (1 April 2026) with earlier versions archived (15 of 30). Inputs stay on the owner's hardware, so no privacy statement applies. ShieldGemma 1 was fine-tuned on synthetic internal data and public datasets described in a technical report, and ShieldGemma 2 on a WebLI subset and synthetic images with CSAM filtering. No data is released (20 of 30). No deprecation policy or dated model retirements (5 of 20). The weights carry no telemetry of their own, with no statement either way, and downloads go through Hugging Face, Kaggle or Ollama (10 of 20). 15 + 20 + 5 + 10 = 50.

### Schema & documentation: ShieldGemma +7

- LlamaFirewall (49): Framework reading. There is no API reference page and no specification. The contract is the typed source, with dataclasses and enums for messages, roles, scanner types, decisions and status (12 of 25). No llms.txt on the docs site (404 again on 4 November). The docs are Markdown in the repository (5 of 10). Each scanner has a page that states its purpose and the risks it covers. None says when not to use it or what it misses (12 of 20). Roles, scanner types and decisions are enums. `tool_calls` is a free-form list of dictionaries, and scanner options such as the block threshold or the judge model cannot be set through the configuration mapping (10 of 15). Eleven example scripts, a notebook and five tutorials. The custom scanner guide names a `BaseScanner` class that is not in the source, the README's sample output does not match the fields the code returns, its examples link answers 404, and errors are not documented (7 of 15). PyPI version numbers only, with no changelog (3 of 15). 12+5+12+10+7+3 = 49.
- ShieldGemma (56): Read for a model the owner serves. There is no specification of its own. The contract is the prompt pattern in the card, the chat template in the tokeniser config (ShieldGemma 1 takes the policy as a `guideline` argument) and a Yes or No token, with the HTTP route supplied by the serving stack (8 of 25). gemma.ai/llms.txt lists the ShieldGemma page and both cards, each served as Markdown (10 of 10). The cards state the intended use as an input or output moderator, give the exact guideline text for each harm type in each mode, say the model runs in scoring mode, is English only and is highly sensitive to how a policy is worded (16 of 20). Fixed harm types and a binary answer, with the policy as free text (10 of 15). A Transformers snippet for each generation and a Keras notebook. No failure modes or error cases are documented (8 of 15). Two named generations with no changelog, and the ai.google.dev page is marked deprecated in favour of gemma.ai (4 of 15). 8 + 10 + 16 + 10 + 8 + 4 = 56.

## Score by category

| Category | Weight | LlamaFirewall | ShieldGemma | Edge |
| --- | --- | --- | --- | --- |
| Reliability | 16% (20 this run) | 48 | 46 | LlamaFirewall +2 |
| Performance | 10%, pending | pending | pending | not scored in this run |
| Schema & documentation | 13% (16.2 this run) | 49 | 56 | ShieldGemma +7 |
| Agent ergonomics | 13% (16.2 this run) | 60 | 72 | ShieldGemma +12 |
| Security & auth | 14% (17.5 this run) | 56 | 53 | LlamaFirewall +3 |
| Payments & pricing | 10% (12.5 this run) | 50 | 52 | ShieldGemma +2 |
| Task success | 10%, pending | pending | pending | not scored in this run |
| Maintenance & community | 7% (8.8 this run) | 15 | 40 | ShieldGemma +25 |
| Transparency & trust | 7% (8.8 this run) | 62 | 54 | LlamaFirewall +8 |
| Negative events | ≤15 | 0 | 0 | |
| **Total** | | **50.1 · D** | **54 · C** | |

## Facts side by side

| Fact | LlamaFirewall | ShieldGemma |
| --- | --- | --- |
| Kind | Agent framework | Model API |
| Vendor | Meta | Google DeepMind |
| Hosted endpoint | no (local only) | no (local only) |
| Transports |  |  |
| Auth | None | None |
| Pricing | Free | Free |
| x402 | no | no |
| Licence | MIT (library). The Prompt Guard 2 weights it downloads are under the Llama 4 Community Licence | Gemma Terms of Use (last modified 1 April 2026) with the Gemma Prohibited Use Policy. Not an OSI licence |
| Read-only variant documented | no | no |
| llms.txt | no | yes |
| Last release | 2025-05-29 | 2025-04-04 |
| Terms last updated | no document linked | 2026-04-01 |
| Privacy policy last updated | no document linked | no document linked |
| Customer content may train models | not read yet | yes |
| Terms restrict automated access | not read yet | not found in the text |
| Terms restrict benchmarking | not read yet | not found in the text |
| Terms or service can change without notice | not read yet | not found in the text |
| Arbitration or class-action waiver | not read yet | not found in the text |
| Popularity | 4.4k stars | none |

## How the scores have moved

### LlamaFirewall

- -1.0 on 4 November 2026: new evidence
- +0.3 on 27 October 2026: checklist

### ShieldGemma

No score change since it was listed.

## Verdicts

**LlamaFirewall.** One `scan()` call runs several checks on the owner's machine and returns a short typed result. The newest version, 1.0.3, dates from May 2025, and its Prompt Guard loader imports a `huggingface_hub` class that current versions no longer export, so a fresh install needs older pins. The classifier weights also need Meta's manual approval.

**ShieldGemma.** Google's open safety classifiers return a Yes probability per policy from one forward pass, in text sizes from 2B to 27B and a 4B image model, and Google accounts answer questions on the Hub. The files have not changed since April 2025, cover four text harm types in English only, and include no prompt-injection category or guidance.

## Before you call either

### LlamaFirewall

1. Pin `huggingface_hub` below 1.0 and a matching `transformers` 4.x before importing the Prompt Guard scanner from the 1.0.3 wheel, or install from main
2. Get access to `meta-llama/Llama-Prompt-Guard-2-86M` and set a Hugging Face token first. Without one the loader prompts for a login and a headless run stalls
3. Call `scan_async` inside a running event loop. `scan()` wraps `asyncio.run` and fails there. `scan_async` returns score 0.0 and reason `default` on every allow
4. Split text longer than 512 tokens yourself before a Prompt Guard scan. The library truncates and does not chunk
5. Do not feed a block `reason` back to the model. The Prompt Guard reason quotes the full scanned text, and the hidden ASCII reason decodes the hidden payload

### ShieldGemma

1. Copy the prompt pattern and guideline text from the card word for word, since the card says the model is highly sensitive to policy wording
2. Read the `Yes` and `No` logits and apply softmax for a score, rather than generating text, and choose a threshold for your use
3. Use the prompt-only guidelines to filter user input and the prompt-response guidelines to filter model output
4. Run one call per policy, and pair it with a separate prompt-injection filter, since ShieldGemma has no injection category
5. Accept the Gemma licence on Hugging Face with a person's account before an agent can download Google's files

## Questions

### Which is better for AI agents, LlamaFirewall or ShieldGemma?

ShieldGemma scores 54 (C) on agent readiness against LlamaFirewall's 50.1 (D), and leads in 4 of 7 scored categories. LlamaFirewall leads on transparency & trust.

### Can an agent call LlamaFirewall and ShieldGemma without installing anything?

No hosted endpoint is listed for LlamaFirewall. No hosted endpoint is listed for ShieldGemma.

### Are LlamaFirewall and ShieldGemma open source?

LlamaFirewall is open source (MIT (library). The Prompt Guard 2 weights it downloads are under the Llama 4 Community Licence). No open-source release is listed for ShieldGemma.


## For agents

- This comparison as JSON: https://www.anchorterminal.com/compare/llamafirewall-vs-shieldgemma.json, and with the fewest tokens: https://www.anchorterminal.com/compare/llamafirewall-vs-shieldgemma.min.md
- Over MCP at https://www.anchorterminal.com/mcp (no key): `compare_tools {"a": "llamafirewall", "b": "shieldgemma"}`. From a terminal: `anchor compare llamafirewall shieldgemma`
- Each listing in full: https://www.anchorterminal.com/api/v1/tools/llamafirewall.json and https://www.anchorterminal.com/api/v1/tools/shieldgemma.json

## Other comparisons with LlamaFirewall or ShieldGemma

- [Amazon Bedrock Guardrails vs ShieldGemma](https://www.anchorterminal.com/compare/amazon-bedrock-guardrails-vs-shieldgemma.md)
- [Azure AI Content Safety (Prompt Shields) vs ShieldGemma](https://www.anchorterminal.com/compare/azure-ai-content-safety-vs-shieldgemma.md)
- [Cisco AI Defense Inspection API vs ShieldGemma](https://www.anchorterminal.com/compare/cisco-ai-defense-inspection-vs-shieldgemma.md)
- [Google Cloud Model Armor vs ShieldGemma](https://www.anchorterminal.com/compare/google-model-armor-vs-shieldgemma.md)
- [Granite Guardian vs ShieldGemma](https://www.anchorterminal.com/compare/granite-guardian-vs-shieldgemma.md)
- [Guardrails AI vs ShieldGemma](https://www.anchorterminal.com/compare/guardrails-ai-vs-shieldgemma.md)
- [Lakera Guard (Check Point AI Guardrails) vs ShieldGemma](https://www.anchorterminal.com/compare/lakera-guard-vs-shieldgemma.md)
- [Llama Guard 4 vs ShieldGemma](https://www.anchorterminal.com/compare/llama-guard-vs-shieldgemma.md)
- [Mistral Moderation API vs ShieldGemma](https://www.anchorterminal.com/compare/mistral-moderation-vs-shieldgemma.md)
- [NVIDIA NeMo Guardrails vs ShieldGemma](https://www.anchorterminal.com/compare/nemo-guardrails-vs-shieldgemma.md)
- [OpenAI Guardrails vs ShieldGemma](https://www.anchorterminal.com/compare/openai-guardrails-vs-shieldgemma.md)
- [OpenAI Moderation API vs ShieldGemma](https://www.anchorterminal.com/compare/openai-moderation-vs-shieldgemma.md)
- [Prisma AIRS AI Runtime Security API vs ShieldGemma](https://www.anchorterminal.com/compare/prisma-airs-vs-shieldgemma.md)
- [Amazon Bedrock Guardrails vs LlamaFirewall](https://www.anchorterminal.com/compare/amazon-bedrock-guardrails-vs-llamafirewall.md)
- [Azure AI Content Safety (Prompt Shields) vs LlamaFirewall](https://www.anchorterminal.com/compare/azure-ai-content-safety-vs-llamafirewall.md)
- [Cisco AI Defense Inspection API vs LlamaFirewall](https://www.anchorterminal.com/compare/cisco-ai-defense-inspection-vs-llamafirewall.md)
- [Google Cloud Model Armor vs LlamaFirewall](https://www.anchorterminal.com/compare/google-model-armor-vs-llamafirewall.md)
- [Granite Guardian vs LlamaFirewall](https://www.anchorterminal.com/compare/granite-guardian-vs-llamafirewall.md)
- [Guardrails AI vs LlamaFirewall](https://www.anchorterminal.com/compare/guardrails-ai-vs-llamafirewall.md)
- [Lakera Guard (Check Point AI Guardrails) vs LlamaFirewall](https://www.anchorterminal.com/compare/lakera-guard-vs-llamafirewall.md)
- [LlamaFirewall vs NVIDIA NeMo Guardrails](https://www.anchorterminal.com/compare/llamafirewall-vs-nemo-guardrails.md)
- [LlamaFirewall vs OpenAI Guardrails](https://www.anchorterminal.com/compare/llamafirewall-vs-openai-guardrails.md)
- [LlamaFirewall vs Prisma AIRS AI Runtime Security API](https://www.anchorterminal.com/compare/llamafirewall-vs-prisma-airs.md)
- [AgentGate vs LlamaFirewall](https://www.anchorterminal.com/compare/agentgate-vs-llamafirewall.md)
- [AgentGate vs ShieldGemma](https://www.anchorterminal.com/compare/agentgate-vs-shieldgemma.md)
- [Llama Guard 4 vs LlamaFirewall](https://www.anchorterminal.com/compare/llama-guard-vs-llamafirewall.md)
- [LlamaFirewall vs Presidio](https://www.anchorterminal.com/compare/llamafirewall-vs-microsoft-presidio.md)
- [LlamaFirewall vs Mistral Moderation API](https://www.anchorterminal.com/compare/llamafirewall-vs-mistral-moderation.md)
- [Presidio vs ShieldGemma](https://www.anchorterminal.com/compare/microsoft-presidio-vs-shieldgemma.md)
- [LlamaFirewall vs OpenAI Moderation API](https://www.anchorterminal.com/compare/llamafirewall-vs-openai-moderation.md)

---

Query this catalogue directly over MCP: https://www.anchorterminal.com/mcp (Streamable HTTP, no key). How: https://www.anchorterminal.com/agents/#mcp