Agent Reliability OSS

Open-source reliability instrumentation for AI agents.

A stable, local-first Python SDK for measuring and testing whether agents meet explicit reliability objectives. No hosted account, API key, or mandatory runtime dependency.

View on GitHub View on PyPI

Install

Reliability primitives, without a platform dependency.

Wrap one logical agent execution, evaluate the relevant outcome, and calculate local reliability against an SLO. The base package sends nothing over the network.

python -m pip install agent-reliability

Language: Python 3.11–3.13
License: Apache-2.0
Runtime dependencies: Zero in base install
Current documented release: 1.2.1

instrument.pyfrom agent_reliability.sdk import AgentReliability

sdk = AgentReliability()

with sdk.run(
    agent_id="support-agent",
    name="Support Agent",
    version="1",
) as run:
    # Run your agent, evaluate its outcome,
    # and record explicit reliability evidence.
    ...

What it provides

Build reliability semantics into the application boundary.

Explicit instrumentation

Sync and async contexts for one logical agent execution, without monkey-patching a framework.

Deterministic evaluation

PASS, FAIL, and UNKNOWN outcomes with evaluator identity, version, configuration, and provenance.

Local reliability

Exact aggregation, SLO evaluation, error budgets, burn rate, and measurement-health tracking.

OpenTelemetry bridge

Join an existing host trace while your application retains ownership of providers and exporters.

Open source SDK

Instrument and measure locally.

The OSS package provides application-side instrumentation, evaluators, provenance, reliability calculations, and test/CI assertions. It works offline and does not require Gentlity.

Read the documentation

Gentlity platform

Operate reliability over time.

The commercial platform adds persisted evidence, historical analysis, journey coverage, revisioned reliability policy, and auditable deployment decisions for teams.

Explore Gentlity

Early access

Bring reliability discipline to your agent stack.

We are working with teams running AI agents in production.

Request Early Access