sentinos (Python SDK)

High-level Python SDK for Sentinos. Use it to connect agent workloads to governed execution, policy-backed decisions, trace forensics, and Chronos context with one environment-driven client.

If you want the lower-level generated API client instead, install sentinos-sdk-core. Most users should start with sentinos.

Install

pip install sentinos

Optional extras:

pip install "sentinos[providers]"  # OpenAI, Anthropic, and Bedrock helpers
pip install "sentinos[otel]"       # OpenTelemetry helpers
pip install "sentinos[langchain]"  # LangChain adapters
pip install "sentinos[grpc]"       # Native gRPC example support

Start here

Set the standard environment variables once:

export SENTINOS_BASE_URL="https://api.sentinos.ai"
export SENTINOS_ORG_ID="<org-id>"
export SENTINOS_ACCESS_TOKEN="<access-token>"

Then create the client and make a governed call:

from sentinos import SentinosClient

client = SentinosClient.from_env()
trace = client.kernel.execute(
    agent_id="assistant-1",
    session_id="sess-1",
    intent={"type": "tool_call", "tool": "example.tool", "args": {"x": 1}},
)

print(trace.decision, trace.trace_id)

If you host services separately, override SENTINOS_KERNEL_URL, SENTINOS_ARBITER_URL, or SENTINOS_CHRONOS_URL instead of switching constructors.

Common workflows

• Govern direct provider calls with LLMGuard
• Replay and export trace evidence for audits
• Simulate Arbiter policy changes before rollout
• Pull Chronos snapshots for context-aware investigation

LLMGuard example

from sentinos import LLMGuard, SentinosClient

client = SentinosClient.from_env()
guard = LLMGuard(kernel=client.kernel, agent_id="assistant-1", session_id="sess-42")

result = guard.run(
    provider="openai",
    operation="chat.completions",
    model="gpt-4o-mini",
    request={"messages": [{"role": "user", "content": "Summarize the incident."}]},
    invoke=lambda: {"id": "resp-1", "model": "gpt-4o-mini"},
)

print(result.trace.trace_id, result.trace.decision)

Trace replay example

from sentinos import SentinosClient

client = SentinosClient.from_env()
replay = client.traces.replay_trace(
    "11111111-1111-1111-1111-111111111111",
    request={"include_explain": True},
)

print(replay.decision, replay.fidelity)

Enterprise auth

Workforce token exchange is available when you need employee-scoped access:

from sentinos import SentinosClient, WorkforceAssertion, WorkforceTokenProvider
from sentinos.auth.jwt import JWTAuth

workforce_provider = WorkforceTokenProvider.from_env(
    assertion_provider=lambda: WorkforceAssertion(
        external_subject="employee-123",
        email="employee@enterprise.example",
        groups=["ai-users"],
    )
)

client = SentinosClient.from_env(auth=JWTAuth(workforce_provider))

Docs & examples

• Sentinos SDK docs
• PyPI package
• Low-level generated client (sentinos-sdk-core)

Included example files:

• examples/quickstart.py
• examples/auth_api_key.py
• examples/protocols/grpc_execute_smoke.py
• docs/cookbooks/quickstart.md
• docs/cookbooks/chronos_snapshots.md
• docs/cookbooks/policy_simulation.md