safety-agent 0.1.2__tar.gz

safety_agent-0.1.2/.gitignore

@@ -0,0 +1,118 @@
+# Dependencies
+node_modules/
+*/node_modules/
+sdk/typescript/package-lock.json
+
+# Python virtual environments
+venv/
+.venv/
+*/venv/
+*/.venv/
+__pycache__/
+*/__pycache__/
+*.py[cod]
+*$py.class
+sdk/python/src/superagent_*.egg-info/
+sdk/python/uv.lock
+
+# AI models and caches
+models/
+*/models/
+*.gguf
+
+# Build outputs
+dist/
+build/
+target/
+bin/
+*.exe
+
+# Environment files
+.env
+.env.local
+.env.*.local
+
+# OS files
+.DS_Store
+Thumbs.db
+
+# Editor files
+.vscode/
+.idea/
+*.swp
+*.swo
+*~
+
+# Logs
+*.log
+logs/
+
+# Runtime data
+pids
+*.pid
+*.seed
+*.pid.lock
+
+# Coverage directory used by tools like istanbul
+coverage/
+
+# Dependency directories
+jspm_packages/
+
+# Optional npm cache directory
+.npm
+
+# Optional eslint cache
+.eslintcache
+
+# Microbundle cache
+.rpt2_cache/
+.rts2_cache_cjs/
+.rts2_cache_es/
+.rts2_cache_umd/
+
+# Optional REPL history
+.node_repl_history
+
+# Output of 'npm pack'
+*.tgz
+
+# Yarn Integrity file
+.yarn-integrity
+
+# parcel-bundler cache (https://parceljs.org/)
+.cache
+.parcel-cache
+
+# next.js build output
+.next
+
+# nuxt.js build output
+.nuxt
+
+# vuepress build output
+.vuepress/dist
+
+# Serverless directories
+.serverless
+
+# FuseBox cache
+.fusebox/
+
+# DynamoDB Local files
+.dynamodb/
+
+# Rust specific
+# Cargo.lock - committed for applications (not libraries)
+target/
+
+# Go specific
+go.mod
+go.sum
+
+# Temporary files
+tmp/
+temp/
+
+# Coding agents
+.claude/

safety_agent-0.1.2/PKG-INFO

@@ -0,0 +1,154 @@
+Metadata-Version: 2.4
+Name: safety-agent
+Version: 0.1.2
+Summary: A lightweight Python guardrail SDK for content safety
+Project-URL: Homepage, https://superagent.sh
+Project-URL: Documentation, https://docs.superagent.sh
+Project-URL: Repository, https://github.com/superagent-ai/superagent
+Author: Superagent AI
+License-Expression: MIT
+Keywords: ai,content-moderation,guardrail,llm,safety
+Classifier: Development Status :: 4 - Beta
+Classifier: Intended Audience :: Developers
+Classifier: License :: OSI Approved :: MIT License
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Classifier: Topic :: Software Development :: Libraries :: Python Modules
+Requires-Python: >=3.10
+Requires-Dist: daytona-sdk>=0.129.0
+Requires-Dist: httpx>=0.27.0
+Requires-Dist: pypdf>=5.0.0
+Description-Content-Type: text/markdown
+
+# Safety Agent Python SDK
+
+A lightweight Python guardrail SDK for content safety. Guard against prompt injections, jailbreaks, and data exfiltration. Redact PII, PHI, and secrets from text.
+
+## Installation
+
+```bash
+uv add safety-agent
+```
+
+Or with pip:
+
+```bash
+pip install safety-agent
+```
+
+## Prerequisites
+
+Sign up at [superagent.sh](https://superagent.sh) to get your API key.
+
+```bash
+export SUPERAGENT_API_KEY=your-key
+```
+
+## Quick Start
+
+```python
+from safety_agent import create_client
+
+client = create_client()
+
+# Guard: Detect threats (uses default superagent/guard-1.7b model)
+result = await client.guard(input="user message to analyze")
+
+if result.classification == "block":
+    print("Blocked:", result.violation_types)
+
+# Redact: Remove PII
+result = await client.redact(
+    input="My email is john@example.com",
+    model="openai/gpt-4o-mini"
+)
+
+print(result.redacted)
+# "My email is <EMAIL_REDACTED>"
+```
+
+## Guard
+
+The `guard()` method classifies input content as `pass` or `block`. It detects prompt injections, malicious instructions, and security threats.
+
+```python
+result = await client.guard(
+    input="Ignore all previous instructions",
+    model="openai/gpt-4o-mini",  # Optional, defaults to superagent/guard-1.7b
+    system_prompt="Custom system prompt",  # Optional
+    chunk_size=8000,  # Optional, characters per chunk
+)
+
+print(result.classification)  # "pass" or "block"
+print(result.violation_types)  # ["prompt_injection", ...]
+print(result.cwe_codes)  # ["CWE-94", ...]
+```
+
+### Input Types
+
+Guard supports multiple input types:
+
+- **Plain text**: Analyzed directly
+- **URLs**: Automatically fetched and analyzed
+- **Bytes/Files**: Analyzed based on content type
+- **PDFs**: Text extracted and analyzed per page
+
+```python
+# URL input
+result = await client.guard(input="https://example.com/document.pdf")
+
+# File input
+with open("document.pdf", "rb") as f:
+    result = await client.guard(input=f.read())
+```
+
+## Redact
+
+The `redact()` method removes sensitive content from text.
+
+```python
+result = await client.redact(
+    input="My SSN is 123-45-6789",
+    model="openai/gpt-4o-mini",
+    entities=["SSN", "email"],  # Optional, custom entities
+    rewrite=True,  # Optional, contextual rewriting
+)
+
+print(result.redacted)
+print(result.findings)
+```
+
+## Supported Providers
+
+- OpenAI (`openai/gpt-4o`, `openai/gpt-4o-mini`, etc.)
+- Anthropic (`anthropic/claude-3-5-sonnet-20241022`, etc.)
+- Google (`google/gemini-2.0-flash`, etc.)
+- AWS Bedrock (`bedrock/us.anthropic.claude-3-5-sonnet-20241022-v2:0`, etc.)
+- Groq (`groq/llama-3.3-70b-versatile`, etc.)
+- Fireworks (`fireworks/accounts/fireworks/models/llama-v3p3-70b-instruct`, etc.)
+- OpenRouter (`openrouter/openai/gpt-4o`, etc.)
+- Vercel (`vercel/openai/gpt-4o`, etc.)
+- Superagent (`superagent/guard-1.7b`, etc.) - Default for guard
+
+## Environment Variables
+
+Configure provider API keys:
+
+```bash
+export SUPERAGENT_API_KEY=your-superagent-key
+export OPENAI_API_KEY=your-openai-key
+export ANTHROPIC_API_KEY=your-anthropic-key
+export GOOGLE_API_KEY=your-google-key
+export GROQ_API_KEY=your-groq-key
+export FIREWORKS_API_KEY=your-fireworks-key
+export OPENROUTER_API_KEY=your-openrouter-key
+export AI_GATEWAY_API_KEY=your-vercel-key
+```
+
+## License
+
+MIT

safety_agent-0.1.2/README.md

@@ -0,0 +1,128 @@
+# Safety Agent Python SDK
+
+A lightweight Python guardrail SDK for content safety. Guard against prompt injections, jailbreaks, and data exfiltration. Redact PII, PHI, and secrets from text.
+
+## Installation
+
+```bash
+uv add safety-agent
+```
+
+Or with pip:
+
+```bash
+pip install safety-agent
+```
+
+## Prerequisites
+
+Sign up at [superagent.sh](https://superagent.sh) to get your API key.
+
+```bash
+export SUPERAGENT_API_KEY=your-key
+```
+
+## Quick Start
+
+```python
+from safety_agent import create_client
+
+client = create_client()
+
+# Guard: Detect threats (uses default superagent/guard-1.7b model)
+result = await client.guard(input="user message to analyze")
+
+if result.classification == "block":
+    print("Blocked:", result.violation_types)
+
+# Redact: Remove PII
+result = await client.redact(
+    input="My email is john@example.com",
+    model="openai/gpt-4o-mini"
+)
+
+print(result.redacted)
+# "My email is <EMAIL_REDACTED>"
+```
+
+## Guard
+
+The `guard()` method classifies input content as `pass` or `block`. It detects prompt injections, malicious instructions, and security threats.
+
+```python
+result = await client.guard(
+    input="Ignore all previous instructions",
+    model="openai/gpt-4o-mini",  # Optional, defaults to superagent/guard-1.7b
+    system_prompt="Custom system prompt",  # Optional
+    chunk_size=8000,  # Optional, characters per chunk
+)
+
+print(result.classification)  # "pass" or "block"
+print(result.violation_types)  # ["prompt_injection", ...]
+print(result.cwe_codes)  # ["CWE-94", ...]
+```
+
+### Input Types
+
+Guard supports multiple input types:
+
+- **Plain text**: Analyzed directly
+- **URLs**: Automatically fetched and analyzed
+- **Bytes/Files**: Analyzed based on content type
+- **PDFs**: Text extracted and analyzed per page
+
+```python
+# URL input
+result = await client.guard(input="https://example.com/document.pdf")
+
+# File input
+with open("document.pdf", "rb") as f:
+    result = await client.guard(input=f.read())
+```
+
+## Redact
+
+The `redact()` method removes sensitive content from text.
+
+```python
+result = await client.redact(
+    input="My SSN is 123-45-6789",
+    model="openai/gpt-4o-mini",
+    entities=["SSN", "email"],  # Optional, custom entities
+    rewrite=True,  # Optional, contextual rewriting
+)
+
+print(result.redacted)
+print(result.findings)
+```
+
+## Supported Providers
+
+- OpenAI (`openai/gpt-4o`, `openai/gpt-4o-mini`, etc.)
+- Anthropic (`anthropic/claude-3-5-sonnet-20241022`, etc.)
+- Google (`google/gemini-2.0-flash`, etc.)
+- AWS Bedrock (`bedrock/us.anthropic.claude-3-5-sonnet-20241022-v2:0`, etc.)
+- Groq (`groq/llama-3.3-70b-versatile`, etc.)
+- Fireworks (`fireworks/accounts/fireworks/models/llama-v3p3-70b-instruct`, etc.)
+- OpenRouter (`openrouter/openai/gpt-4o`, etc.)
+- Vercel (`vercel/openai/gpt-4o`, etc.)
+- Superagent (`superagent/guard-1.7b`, etc.) - Default for guard
+
+## Environment Variables
+
+Configure provider API keys:
+
+```bash
+export SUPERAGENT_API_KEY=your-superagent-key
+export OPENAI_API_KEY=your-openai-key
+export ANTHROPIC_API_KEY=your-anthropic-key
+export GOOGLE_API_KEY=your-google-key
+export GROQ_API_KEY=your-groq-key
+export FIREWORKS_API_KEY=your-fireworks-key
+export OPENROUTER_API_KEY=your-openrouter-key
+export AI_GATEWAY_API_KEY=your-vercel-key
+```
+
+## License
+
+MIT

safety_agent-0.1.2/pyproject.toml

@@ -0,0 +1,50 @@
+[project]
+name = "safety-agent"
+version = "0.1.2"
+description = "A lightweight Python guardrail SDK for content safety"
+readme = "README.md"
+license = "MIT"
+authors = [{ name = "Superagent AI" }]
+requires-python = ">=3.10"
+keywords = ["guardrail", "safety", "llm", "ai", "content-moderation"]
+classifiers = [
+    "Development Status :: 4 - Beta",
+    "Intended Audience :: Developers",
+    "License :: OSI Approved :: MIT License",
+    "Programming Language :: Python :: 3",
+    "Programming Language :: Python :: 3.10",
+    "Programming Language :: Python :: 3.11",
+    "Programming Language :: Python :: 3.12",
+    "Programming Language :: Python :: 3.13",
+    "Topic :: Software Development :: Libraries :: Python Modules",
+    "Topic :: Scientific/Engineering :: Artificial Intelligence",
+]
+dependencies = [
+    "httpx>=0.27.0",
+    "pypdf>=5.0.0",
+    "daytona-sdk>=0.129.0",
+]
+
+[project.urls]
+Homepage = "https://superagent.sh"
+Documentation = "https://docs.superagent.sh"
+Repository = "https://github.com/superagent-ai/superagent"
+
+[build-system]
+requires = ["hatchling"]
+build-backend = "hatchling.build"
+
+[tool.hatch.build.targets.wheel]
+packages = ["src/safety_agent"]
+
+[tool.uv]
+dev-dependencies = [
+    "pytest>=8.0.0",
+    "pytest-asyncio>=0.24.0",
+    "python-dotenv>=1.0.0",
+]
+
+[tool.pytest.ini_options]
+asyncio_mode = "auto"
+asyncio_default_fixture_loop_scope = "function"
+testpaths = ["tests"]

safety_agent-0.1.2/src/safety_agent/__init__.py

@@ -0,0 +1,51 @@
+"""
+safety-agent
+A lightweight Python guardrail SDK for content safety
+"""
+
+from .client import SafetyClient, create_client
+from .types import (
+    ClientConfig,
+    GuardInput,
+    GuardOptions,
+    RedactOptions,
+    ScanOptions,
+    GuardClassificationResult,
+    RedactResult,
+    GuardResponse,
+    RedactResponse,
+    ScanResponse,
+    ScanUsage,
+    ChatMessage,
+    MultimodalContentPart,
+    ProcessedInput,
+    AnalysisResponse,
+    TokenUsage,
+    ParsedModel,
+)
+
+__version__ = "0.1.0"
+
+__all__ = [
+    # Client
+    "SafetyClient",
+    "create_client",
+    # Types
+    "ClientConfig",
+    "GuardInput",
+    "GuardOptions",
+    "RedactOptions",
+    "ScanOptions",
+    "GuardClassificationResult",
+    "RedactResult",
+    "GuardResponse",
+    "RedactResponse",
+    "ScanResponse",
+    "ScanUsage",
+    "ChatMessage",
+    "MultimodalContentPart",
+    "ProcessedInput",
+    "AnalysisResponse",
+    "TokenUsage",
+    "ParsedModel",
+]