gaas-pydantic-ai

GaaS (Governance as a Service) integration for Pydantic AI (v2).

Every tool call is governed before it executes: the toolset submits a governance intent to GaaS, and only an APPROVE (or an approved escalation) lets the tool run. On BLOCK, GovernanceBlockedError is raised — and because Pydantic AI propagates ordinary exceptions out of agent.run() natively, a blocked action halts the run. No wrappers fighting the framework: this is the framework's own semantics.

pip install gaas-pydantic-ai[pydantic-ai]

Quickstart

from pydantic_ai import Agent
from gaas_pydantic_ai import govern_tools, GaaSGovernanceConfig

def search_web(query: str) -> str:
    """Search the web."""
    ...

def send_email(to: str, subject: str, body: str) -> str:
    """Send an email."""
    ...

config = GaaSGovernanceConfig(api_key="gsk_...", agent_id="my-agent")

agent = Agent(
    "anthropic:claude-sonnet-5",
    toolsets=[govern_tools([search_web, send_email], config=config)],
)

result = await agent.run("Email the Q3 report to finance@acme.com")

Already have a toolset? Wrap it directly:

from gaas_pydantic_ai import govern_toolset

agent = Agent(model, toolsets=[govern_toolset(my_toolset, config=config)])

Verdict flow

tool call → GaaS intent → ┌─────────┐
                          │ APPROVE │ → tool executes
                          │ BLOCK   │ → GovernanceBlockedError aborts agent.run()
                          │ ESCALATE│ → aborts by default; hold-and-poll optional
                          └─────────┘

GovernanceBlockedError is never converted into model-visible text — the adapter deliberately does not use ModelRetry, so the model cannot route around a block.

Configuration

config = GaaSGovernanceConfig(
    api_url="https://api.gaas.is",    # GaaS API endpoint
    api_key="gsk_...",                # Your API key
    agent_id="my-agent",              # Appears in the audit trail
    block_on_escalate=True,           # Raise on ESCALATE (default)
    timeout_seconds=5.0,              # Governance check timeout (fail open)
    sensitivity="INTERNAL",           # Default sensitivity for tool inputs
    raise_on_governance_error=False,  # Fail open if GaaS is unreachable
    extra_regulatory_domains=["HIPAA"],
    extra_data_categories=["PHI"],
    hold_on_escalate=False,           # Wait for the human decision on ESCALATE
    escalation_poll_seconds=5.0,
    escalation_max_wait_seconds=600.0,
)

Hold-and-Poll on ESCALATE

With hold_on_escalate=True, an ESCALATE verdict holds the tool call while GaaS routes the escalation to a human reviewer: approve/modify lets the tool execute; deny raises GovernanceBlockedError (ESCALATE_DENY); timeout raises ESCALATE_TIMEOUT.

Handling blocked runs

from gaas_pydantic_ai import GovernanceBlockedError

try:
    result = await agent.run("wire $250k to the new vendor")
except GovernanceBlockedError as err:
    print(err.verdict)                  # BLOCK / ESCALATE / ESCALATE_DENY / ESCALATE_TIMEOUT
    print(err.decision_id)              # audit reference
    print(err.blocking_policies)        # policy IDs that triggered the block
    print(err.governance_proof_token)   # GPT token ID for the legal audit trail

Links

• Docs: https://gaas.to/sdks.html
• GaaS: https://gaas.is