Skip to main content
Trinitite SDK

One developer layer for the full AI lifecycle.

Govern the model you already use. Test the behavior that matters. Explain each decision. Improve what fails. Keep operations, risk, and evidence in the same typed foundation.
Start with Python or TypeScriptRead the SDK model

Status: Stable · GA Companion docs: Get started · Concepts · Verdicts and modes · Evals onboarding Source of truth: the shipped Python and TypeScript SDKs plus their backing API contracts.

Your first integration is the same idea in both SDKs​

Set a baseline. Wrap a familiar provider call. Read a GovernanceResult. Python and TypeScript are live, first-class SDKs, and the language tabs stay in sync across the docs.

The core SDK path

A familiar model call with a typed decision beside it.

Python / TypeScript
app.py
from trinitite import Trinitite

tr = Trinitite()
tr.govern(frameworks=["soc2"], mode="enforce")

response = tr.client(
  "openai",
  credential="cred_openai_prod",
).chat.completions.create(
  model="gpt-4o",
  messages=[{"role": "user", "content": "Refund order 4821."}],
)

result = tr.result(response)
print(result.verdict)
print(result.violations)
print(result.compliance)

What the foundation gives you​

One integration, a wider runway

Build, prove, and improve in the same loop

01

Connect

Models, agents, MCP, CLI, and Skills share one foundation.

02

Test

Run evals and red-team the behavior that matters to you.

03

Explain

See the verdict, findings, controls, risk, and decision trace.

04

Improve

Turn recurring failures into prompt candidates and better Guardians.

05

Operate

Trace live behavior, replay decisions, and route evidence to your tools.

06

Train and serve

Version focused models, compare them, promote them, or roll back.

The SDK is more than a model wrapper:

  • Model and agent runtime: govern calls and normalize each result.
  • MCP and CLI: apply rules around tool calls and command execution.
  • Skills: bring skill content into the governed asset model.
  • Evals and red-teaming: test traces, captured traffic, or simulated users.
  • Prompt improvement: generate and compare candidates against frozen cases.
  • Training and serving: version focused models, test challengers, promote, route, and roll back.
  • Observability: search, replay, compare, stream, and export decisions.
  • Governance and compliance: connect findings to policies, controls, risks, evidence, and reports.
  • Deployment choice: use SaaS, hybrid, or self-hosted enterprise shapes.

Each surface has its own typed handle. The common posture and result model keep them from becoming separate islands.

Three layers keep a large system simple​

The three-layer model, three nouns, one rule

Scope

A goal

Declare the goal once. Every asset bound to this scope shares the goal and its stacked policies.

tr.scope("refund-flow", goal="Resolve refunds within policy")

The baseline is the shared posture. A scope groups work around one goal. An asset is a model, MCP server, CLI, Skill, connector, sandbox, or RAG system. Assets inherit the baseline and, when scoped, the goal-specific rules.

The value grows with each real call​

Technical workshop

Inspect the closed loop

Nothing advances until you do.

Step 1 of 5

Pin the baseline, goal, and scope.

The loop gives developers a direct path from production behavior to the next change. Evals show where to focus. Explanations show why. Prompt and training work can target repeat failures. Release gates compare the new version against fixed evidence before it ships.

Choose your next move​

Last updated on