AI security for technology companies and SaaS teams
Give your agents identities, check their tool calls against policy, and test them in CI. The Python SDK has adapters for the frameworks most teams use.
Challenge and solution
Challenge
Technology companies shipping AI agents need clear control over which identities can request which tool actions, how those requests are reviewed, and whether a model or prompt change made the app easier to attack.
Solution
The Python SDK (pip install aynigma) lets an integrated caller register an agent identity and submit a tool call for policy evaluation. Adapters cover LangChain, LangGraph, the OpenAI Agents SDK, CrewAI, the Claude Agent SDK, Google ADK and Semantic Kernel. Only requests routed through the integration are evaluated. Your executor keeps the credentials and performs the authorized action. Red team results can run in CI as SARIF or through a GitHub Action.
Key capabilities
Agent Identity
Register agent identities and issue credentials with expiry and overlapping rotation for authorization API access.
Tool-Call Policies
Policy-as-code for tool and argument constraints, with a simulator to test a change before it goes live.
Human Review
Require approval for configured high-risk requests, delivered by email, Slack or Teams.
Audit Trail
Review hash-chained authorization events, and check them offline with aynigma-audit-verify.
SDK adapters
LangChain, LangGraph, OpenAI Agents SDK, CrewAI, Claude Agent SDK, Google ADK and Semantic Kernel, installed as extras of the aynigma package.
CI red teaming
Run red team attacks on a schedule or on each change. Results export as SARIF 2.1.0, run from a GitHub Action, and publish to GitHub or Jira.
What you get
Explicit Authorization
Submit tool-call requests to a policy decision point before a customer-controlled executor acts.
Regressions Caught Before Release
Scheduled and CI red team runs show when a change made an attack work.
Review Context
Keep decision, approval, and actor metadata for follow-up.
Integration Review
Confirm SDK and API fit with your agent framework before beginning a pilot.
Evaluate agent identity and authorization for your stack.
Contact Aynigma to confirm current SDK support, integration scope, and deployment requirements.