Skip to the content
Know your tech before you buy it. Makers: get listed from $49/mo

Augustus

github.com

  • Free plan
  • Windows
  • macOS
  • Linux
  • Self-hosted
7.6 Club score Visit github.com

What Augustus is

Augustus is a self-hosted, Go-based tool for security professionals who assess large language models. It runs adversarial tests against models and produces vulnerability reports, with more than 210 probes across 47 attack categories. These include jailbreaks, prompt injection, data extraction, safety benchmarks and agent attacks. Augustus supports 28 provider categories and 43 generator variants, including hosted providers, Ollama and custom REST endpoints. Its detector set includes more than 90 options, such as pattern matching, LLM-as-a-judge, HarmJudge, Perspective API and unsafe-content detection. Seven prompt transformations include encoding, paraphrase, poetry, translation into low-resource languages and case changes. Results can be displayed as tables or exported as JSON, JSONL or HTML. The project lists concurrent scans, rate limiting, retries and timeout handling. Augustus is free and licensed under Apache 2.0. It compiles to a portable single binary and can be installed with Go or built from source; the README requires Go 1.27.0 or later. Use it only to test systems you own or have explicit permission to assess.

Who it suits

Augustus is aimed at security professionals conducting authorized model testing. It suits teams that can run self-hosted tools and meet the stated Go version requirement.

Good to know, and what to watch

  • More than 210 probes across 47 categories
  • Supports 28 provider categories
  • Exports results as tables, JSON, JSONL or HTML
  • Free and licensed under Apache 2.0
  • Requires Go 1.27.0 or later
  • Designed for authorized security testing only
  • Provider API usage may require credentials

Augustus

Free

Free plan

Getting started

  1. Open the Augustus GitHub repository
  2. Install Go 1.27.0 or later
  3. Install Augustus with Go or build it from source
  4. Compile the portable single binary
  5. Configure an LLM provider or custom REST endpoint
  6. Run scans only against systems you own or are authorized to test
  7. Export results as tables, JSON, JSONL, or HTML

Limits to check

API usage may require credentials from the selected provider.

The Club's review

Augustus is a self-hosted Go scanner for adversarial testing of large language models and AI applications. It is aimed at security professionals and engineering teams conducting authorized assessments across multiple providers. The principal trade-off is clear: unusually broad open-source coverage at no license cost, but more setup and provider configuration than a hosted service.

Overview

Augustus concentrates on finding model vulnerabilities rather than serving as a general-purpose evaluation suite. It tests jailbreak resistance, prompt-injection handling, data exposure, unsafe responses and agent-oriented weaknesses, then produces structured reports. The Apache 2.0 license and portable single-binary distribution suit organizations that want to run or modify the scanner inside their own infrastructure.

Self-hosting keeps scan data in your environment and avoids a subscription, but installation is not entirely turnkey. You install through Go or build from source, and the project requires Go 1.27.0 or later. Provider credentials may also be needed, so teams must budget for the API usage generated by their own model accounts. Augustus is explicitly for systems you own or have permission to test; it should not be used for unsolicited probing.

Key features

The scanner offers more than 210 vulnerability probes across 47 attack categories. Coverage includes jailbreaks, prompt injection, data extraction, safety benchmarks and agent attacks, giving a security review substantially more breadth than a single attack-family check. The breadth is valuable for comparative assessments, although probe count alone does not indicate how a particular model will perform.

Augustus connects to 28 provider categories through 43 generator variants. Named integrations include OpenAI, Anthropic, Azure OpenAI, AWS Bedrock, Vertex AI, Ollama and custom REST endpoints. That range makes it practical for mixed estates and internal endpoints, provided your team can supply and maintain the relevant access.

More than 90 detectors interpret results using approaches such as pattern matching, LLM-as-a-judge, HarmJudge, Perspective API and unsafe-content detection. Seven prompt transformations—including encoding, paraphrase, poetry, low-resource-language translation and case changes—vary the inputs so a model is not assessed against one fixed wording.

Results can be emitted as tables, JSON, JSONL or HTML. Those formats cover interactive review, machine processing and report sharing. Concurrent scanning, rate limiting, retries and timeout handling are listed as production features, which is useful for repeatable jobs and CI/CD pipelines rather than occasional manual experiments. Custom test cases and detection rules, plus GitHub and GitLab support, allow teams to adapt scans to their own applications and pull-request workflows.

Pricing

Augustus — free. The free plan is the Apache 2.0-licensed open-source software itself, with no subscription charge. It is the right fit for security teams that can operate self-hosted tooling and are comfortable paying any separate model-provider API costs. There are no published seat quotas, scan allowances or trial and renewal terms to weigh against a paid tier; the practical cost is infrastructure, setup time and the credentials used for scanning.

Platforms

Augustus supports Linux, macOS and Windows, with self-hosted deployment. Its single-binary form is convenient for controlled environments, while the Go 1.27.0-or-later requirement may exclude teams that cannot standardize on that toolchain. CI/CD and pull-request scanning make it suitable for engineering pipelines, not only standalone analyst workstations.

Who it's for

Choose Augustus if you are a security professional or platform team running authorized tests on several LLM providers and need broad adversarial coverage, automation and exportable findings without a license fee. It is especially appropriate for organizations that already manage internal runners, credentials and CI systems.

Look elsewhere if you need a hosted interface with minimal installation, or if your organization cannot meet the Go and provider-access requirements. The tool's authorization rule also makes it unsuitable for testing systems without explicit permission.

Pros and cons

Pros

  • Wide attack surface coverage: 210-plus probes across 47 categories cover jailbreak, injection, extraction and agent scenarios in one scanner.
  • Broad provider reach: 28 categories, 43 generators and custom REST support accommodate varied model estates.
  • Flexible evidence handling: Table, JSON, JSONL and HTML output serves analysts, pipelines and formal reports.
  • Operational controls: Concurrency, rate limiting, retries and timeouts support repeatable production scans.
  • No software license fee: The Apache 2.0 open-source plan is accessible to teams able to self-host.

Cons

  • Setup burden: Go 1.27.0 or later and source or Go-based installation require developer-tooling support.
  • Usage costs remain possible: Scans may consume paid provider API quotas even though Augustus itself is free.
  • Self-hosted responsibility: Your team must run the scanner, manage credentials and interpret findings rather than relying on a hosted vendor service.

Alternatives

OpenSecureAI Scanner is a better choice when you want a freemium hosted option; its Pro plan costs $49 a month and includes 50,000 API requests, hosted remediation and priority email support.

AIRTA Red Team suits users seeking a free open-source tool that runs through API, Linux, macOS or web access with their own LLM API key.

PyRIT is preferable if a Python-based open-source framework fits your environment and you can configure the required AI endpoints.

Basilisk offers free authorized-use testing through CLI, desktop, Docker or source distribution.

Promptfoo is worth choosing for a freemium community plan with 10,000 red-team probes per month and local or self-hosted operation.

PromptGuard is an alternative for teams wanting hosted scan quotas, including a free tier with 20,000 scans per month.

F5 BIG-IP APM is a paid platform with a free trial for organizations that need an enterprise access-management product rather than a dedicated open-source LLM scanner.

Project Moonshot is another free option for teams evaluating AI security across hosted and self-hosted environments.

For broader discovery, see AI Security Testing Tools and Secrets Scanning Software.

Verdict

Augustus is a strong fit for security teams that want extensive adversarial LLM testing, provider flexibility and CI-ready reports in a free Apache 2.0 package. Choose it when self-hosting and Go-based setup are acceptable; choose a hosted or more specialized alternative when minimizing operational work matters more than breadth and control.

The verdict

Augustus offers broad adversarial test coverage and several report formats in a free, self-hosted package. Check provider access and the Go requirement before setting it up.

Questions people ask

What does Augustus test?

It tests language models with probes for jailbreaks, prompt injection, data extraction, safety benchmarks, agent attacks, and other categories.

Which model providers are supported?

Augustus supports 28 provider categories and 43 generator variants, including OpenAI, Anthropic, Azure OpenAI, AWS Bedrock, Vertex AI, Ollama, and custom REST endpoints.

How can reports be exported?

Scan results can be produced as tables, JSON, JSONL, or HTML reports.

Is Augustus open source?

Yes. The repository identifies it as Apache 2.0 licensed open-source software, available free of charge.

Can it run in CI/CD?

Yes. The listed capabilities include CI/CD scanning, pull-request scanning, custom test cases, and custom detection rules.

What the maker publishes

Attack coverage
The README describes 210+ vulnerability probes across 47 attack categories, including jailbreaks, prompt injection, data extraction, safety benchmarks, and agent attacks. source
Authorized use
The project says Augustus is designed for authorized security testing and instructs users to test only systems they own or have explicit permission to test. source
Detectors
It includes 90+ detectors such as pattern matching, LLM-as-a-judge, HarmJudge, Perspective API, and unsafe content detection. source
Distribution
Augustus compiles as a portable single binary and can be installed with Go or built from source; the README requires Go 1.27.0 or later. source
Founded
2010 source
Founded
2017 source
Headquarters
Austin, Texas, United States source
Intended audience
The README describes Augustus as built for production security testing and identifies security professionals as its users. source
License
The repository identifies Augustus as licensed under Apache 2.0. source
Production features
The project lists concurrent scanning, rate limiting, retry logic, and timeout handling as built-in production features. source
Prompt transformations
Seven prompt transformations include encoding, paraphrase, poetry, low-resource language translation, and case transforms. source
Providers
It supports 28 LLM provider categories with 43 generator variants, including OpenAI, Anthropic, Azure OpenAI, AWS Bedrock, Vertex AI, Ollama, and custom REST endpoints. source
Purpose
Augustus is a Go-based LLM vulnerability scanner for security professionals that tests models against adversarial attacks and produces vulnerability reports. source
Reports
Scan results can be output as tables, JSON, JSONL, or HTML reports. source
Security reporting
The security policy asks researchers to report vulnerabilities privately by email and says receipt will be acknowledged within 48 hours and severity assessed within 7 days. source
Support
The README directs users to open GitHub issues for bugs or feature requests and invites contributions of probes, detectors, and provider integrations. source

Checked against the maker's pages on 4 Oct 2026.