How Dome Works
Dome wraps your agent with configurable guardrails:
| Component | Purpose |
|---|---|
| Guardrail | Pipeline of guards (input or output) |
| Guard | Group of detectors of one type |
| Detector | Individual detection method |
Protection Types
Security Guards
Detect and block adversarial attacks:| Detector | What It Catches |
|---|---|
prompt-injection-mbert | Injected instructions in user input |
prompt-injection-deberta-v3-base | Advanced injection attempts |
encoding-heuristics | Base64, Unicode, and encoding attacks |
security-embeddings | Semantic similarity to known attacks |
Moderation Guards
Filter harmful and inappropriate content:| Detector | What It Catches |
|---|---|
moderation-flashtext | Fast keyword-based toxicity |
moderation-deberta | Neural toxicity classification |
moderations-oai-api | OpenAI Moderation API |
moderation-llamaguard | Llama Guard safety model |
Privacy Guards
Prevent exposure of sensitive data:| Detector | What It Catches |
|---|---|
privacy-presidio | PII (names, emails, SSN, etc.) |
detect-secrets | API keys, passwords, credentials |
Quick Start
You can protect your agents with default guards. The default configuration includes:- Input: Prompt injection detection, encoding heuristics, moderation
- Output: Moderation, PII detection
Configuration Sources
You can pull configurations securely from your registered Vijil agent. Alternatively, you can define configurations using dictionaries or TOML files.Scan Results
Every scan process returns a comprehensive result object describing whether the content is safe, the safe fallback message if flagged, and the execution trace. When content is flagged:- A signal specifies the content was rejected (is_safe becomes false)
- It provides a safe fallback message
- The trace shows which detector flagged it and why
Framework Integrations
Dome is designed to be integrable with popular frameworks and runtimes. You can see the specific framework developer guides for integration patterns during the developer access phase.Performance Options
You can configure protection controls to stop processing when the first guard flags content, or run guards concurrently for maximum efficiency and parallel execution.Work in Progress
The programmatic protection capabilities are currently in private preview and subject to change.
Next Steps
Configuring Guardrails
Detailed guard configuration options
Using Guardrails
Runtime patterns and best practices
Custom Detectors
Build your own detection methods
Observability
Monitoring and tracing setup