vijil-dome 1.0.4

No project description provided

Vijil Dome

Vijil Dome is a fast, lightweight, and highly configurable library for adding runtime guardrails to your AI agents. It combines top open-source LLM safety tools with proprietary Vijil models to detect and respond to unsafe content — with built-in support for observability, tracing, and popular agent frameworks.

🚀 Installation

Install the core library:

pip install vijil-dome

Optional extras for common integrations:

• opentelemetry – OTel-compatible tracing/logging
• google – GCP-native metrics and logging
• langchain – Seamless integration with LangChain/LangGraph
• embeddings – Fast similarity search using annoy

⚠️ Note: annoy is not currently compatible with agents built using Google ADK + Cloud Run. Use in-memory embeddings in those cases.

🔒 Guarding Agents in One Line

from vijil_dome import Dome

dome = Dome()

query = "How can I rob a bank?"
input_scan = dome.guard_input(query)
print(input_scan.is_safe(), input_scan.guarded_response())

# Get a response from your agent 

response = "Here's how to rob a bank!"
output_scan = dome.guard_output(response)
print(output_scan.is_safe(), output_scan.guarded_response())

By default, Dome:

• Scans inputs for prompt injections, jailbreaks, and toxicity
• Scans outputs for toxicity and masks PII

⚙️ Configuration Options

You can configure Dome using a TOML file or a Python dictionary.

Example TOML

[guardrail]
input-guards = ["prompt-injection", "input-toxicity"]
output-guards = ["output-toxicity"]
input-early-exit = false

[prompt-injection]
type = "security"
early-exit = false
methods = ["prompt-injection-deberta-v3-base", "security-llm"]

[prompt-injection.security-llm]
model_name = "gpt-4o"

[input-toxicity]
type = "moderation"
methods = ["moderations-oai-api"]

[output-toxicity]
type = "moderation"
methods = ["moderation-prompt-engineering"]

Same Configuration in Python

config = {
    "input-guards": ["prompt-injection", "input-toxicity"],
    "output-guards": ["output-toxicity"],
    "input-early-exit": False,
    "prompt-injection": {
        "type": "security",
        "early-exit": False,
        "methods": ["prompt-injection-deberta-v3-base", "security-llm"],
        "security-llm": {
            "model_name": "gpt-4o"
        }
    },
    "input_toxicity": {
        "type": "moderation",
        "methods": ["moderations-oai-api"]
    },
    "output_toxicity": {
        "type": "moderation",
        "methods": ["moderation-prompt-engineering"]
    },
}

Dome includes nearly 20 prebuilt guardrails and supports building your own!

👉 For the full list of guardrail methods, advanced config options, and extensibility, check out the Docs.

🔌 Compatibility

Dome works with any agent framework or LLM — it operates directly on strings, so there's no dependency on your stack!

For popular frameworks, we provide dedicated wrappers and tutorials to make integration seamless:

• Google ADK
• LangChain & LangGraph

Observability Integrations:

Dome is compatible with the following observability framworks out of the box

• OpenTelemetry
• Weave (Weights & Biases)
• AgentOps
• Google Cloud Trace

See the documentation for more details

📚 Learn More

Get detailed guides, examples, and custom guardrail walkthroughs in the official documentation →

Have more questions, or want us to include another guardrailing technique? Reach out to us at [e-mail]!