sorin 0.1.9

Python SDK for Sorin — AI agent governance and security

sorin

Python SDK for the Sorin agent authorization and observability platform.

Install

pip install sorin

For local development:

pip install -e ./python

Usage

LLM Inference (Anthropic)

from sorin import SorinLLM

client = SorinLLM(agent_key="your-agent-key")

response = client.messages.create(
    model="claude-sonnet-4-20250514",
    max_tokens=100,
    messages=[{"role": "user", "content": "Hello"}]
)

LLM Inference (OpenAI)

from sorin import SorinOpenAI

client = SorinOpenAI(agent_key="your-agent-key")

response = client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": "Hello"}]
)

Agent Actions (GitHub, AWS, etc.)

from sorin import SorinClient

sorin = SorinClient(agent_key="your-agent-key")

# GitHub
file = sorin.github.read_file("your-org", "your-repo", "README.md")
sorin.github.create_branch("your-org", "your-repo", "my-feature")
sorin.github.push_file("your-org", "your-repo", "notes.md", "# Hello", "add notes", "my-feature")
pr = sorin.github.create_pr("your-org", "your-repo", "My PR", "body", "my-feature")

# AWS — S3
buckets = sorin.aws.s3_list_buckets()
objects = sorin.aws.s3_list_objects("my-bucket", prefix="logs/")
obj = sorin.aws.s3_get_object("my-bucket", "path/to/file.json")
sorin.aws.s3_put_object("my-bucket", "output/report.txt", content="done")
sorin.aws.s3_delete_object("my-bucket", "tmp/scratch.txt")

# AWS — EC2
instances = sorin.aws.ec2_describe_instances()
sorin.aws.ec2_start_instance("i-1234567890abcdef0")
sorin.aws.ec2_stop_instance("i-1234567890abcdef0")

# AWS — Lambda
functions = sorin.aws.lambda_list_functions()
result = sorin.aws.lambda_invoke("my-function", payload='{"key": "value"}')

Write actions and approval-gated permissions are handled automatically — if a 202 is returned the SDK polls until the request is approved or denied, then retries.

Tool calls grouped by LLM turn

The Activity dashboard groups your agent's tool calls under the LLM turn that triggered them. No setup required — the SDK handles it when you use SorinLLM/SorinOpenAI and SorinClient together.

If a single LLM turn fires multiple parallel tool calls, opt into finer-grained labeling by passing the block id:

response = sorin_llm.messages.create(model=..., messages=..., tools=...)
for block in response.content:
    if block.type == "tool_use":
        sorin.github.push_file(**block.input, tool_use_id=block.id)
        # or: sorin.aws.s3_put_object(**block.input, tool_use_id=block.id)

tool_use_id is optional on every sorin.github.* and sorin.aws.* method.

MCP Server

The package ships a sorin CLI that connects any MCP-compatible AI coding tool to the Sorin MCP server.

Claude Code — registers the server automatically:

sorin mcp install --key <your-sorin-agent-key>

Cursor, Windsurf, VS Code, or any host that uses mcpServers JSON config — prints the config block to paste in:

sorin mcp install --key <your-sorin-agent-key> --json

Config file locations:

• Cursor: ~/.cursor/mcp.json
• Windsurf: ~/.codeium/windsurf/mcp_config.json
• VS Code: .vscode/mcp.json (workspace) or user settings

Restart your editor after installing to connect.

Links

• trysorin.com
• Documentation
• PyPI
• GitHub
• Dashboard