pydanticai-tealtiger

Deterministic governance guard for Pydantic AI agents — policy enforcement, PII detection, cost tracking, kill switch, and structured audit evidence.

No LLM in the governance path. All policy evaluation is deterministic, adding <2ms latency per call.

Installation

pip install pydanticai-tealtiger

Quick Start

from pydantic_ai import Agent
from pydanticai_tealtiger import TealTigerGuard

# Zero-config: observe, track cost, detect PII, allow all
guard = TealTigerGuard()

agent = Agent('openai:gpt-4', deps_type=TealTigerGuard)

@agent.tool
async def search(ctx, query: str) -> str:
    # Governance check before execution
    ctx.deps.evaluate(tool="search", args={"query": query})
    result = perform_search(query)
    # Record cost after execution
    ctx.deps.post_call(tool_name="search", result=result)
    return result

# Run with governance
result = await agent.run("Find AI safety papers", deps=guard)

# Inspect governance data
print(guard.cumulative_cost)
print(guard.audit_trail)
print(guard.summary)

Features

Governance Modes

Usage Patterns

Policy Mode with TealEngine

from tealtiger import TealEngine
from pydanticai_tealtiger import TealTigerGuard

engine = TealEngine(policies=[
    {"type": "cost_limit", "max_per_session": 5.00},
    {"type": "pii_block", "action": "DENY"},
])

guard = TealTigerGuard(engine=engine, mode="ENFORCE")

Tool Allowlist

guard = TealTigerGuard(
    mode="ENFORCE",
    tool_allowlist=["search", "compute", "read_file"],
)

Budget Limit

guard = TealTigerGuard(
    mode="ENFORCE",
    budget_limit=1.00,  # $1.00 max per session
)

Kill Switch

guard = TealTigerGuard(mode="ENFORCE")

# Emergency stop — blocks all subsequent calls
guard.freeze()

# Resume operations
guard.unfreeze()

pre_call / post_call Pattern

@agent.tool
async def my_tool(ctx, arg: str) -> str:
    # pre_call is an alias for evaluate
    ctx.deps.pre_call(tool_name="my_tool", args={"arg": arg})

    result = do_work(arg)

    # Record actual cost
    ctx.deps.post_call(
        tool_name="my_tool",
        result=result,
        token_usage={"total_tokens": 500},
    )
    return result

API Reference

TealTigerGuard

Methods

Properties

Development

cd packages/pydanticai-tealtiger
pip install -e ".[dev]"
pytest

License

Apache-2.0 — see LICENSE.