Sierra

Agents reach the customer's systems of record through integrations built on the Agent SDK, executing end-to-end transactions including card and ACH payments, returns, authentication and cancellation saves. Those system interactions are held deterministic and policy-bound. Integrations are engineered through the SDK rather than drawn from a published connector catalog, so breadth depends on the build.

Sourcesierra.ai/product/agent-sdkread 2026-09-05

Agents are composed by mixing and matching skills such as triage, respond and confirm into multi-step workflows executed for a specific use case. The degree of flexibility is set per workflow, so a customer dials between deterministic execution and open-ended reasoning on the same agent.

Sourcesierra.ai/product/agent-sdkread 2026-09-05

Agents ground on the customer's own content through retrieval built into Agent OS. The Agent Data Platform unifies that knowledge alongside customer context across systems, so answers draw on a maintained store rather than a per-conversation fetch. The ingestion and indexing mechanism is not documented publicly, so the maintained store is an inference from the platform layer's stated role.

Sourcesierra.ai/product/agent-sdkread 2026-09-05

Sierra wraps its models in supervisory layers that check outputs against scope and policy before the agent responds, to reduce hallucination and prevent abuse. Access to systems of record is deterministic and controlled, so the customer's policies and security procedures are enforced outside the model rather than trusted to it. Workflow-level flexibility settings bound how far the agent may improvise.

Sourcesierra.ai/product/trust-and-reliabilityread 2026-09-05

Sierra publishes commitments to SOC 2, ISO 27001, ISO 42001, PCI, HIPAA, GDPR, CCPA, CSA STAR and FedRAMP High. A trust portal at trust.sierra.ai carries the specific controls and policies, with documentation available on request. Sierra also states that customer data is never used to train models and that access to systems of record is deterministic and policy-bound rather than left to the model.

Sourcesierra.ai/product/trust-and-reliabilityread 2026-09-05

The platform lets a developer understand and adjust agent behavior by inspecting API calls and logic traces, so the path a conversation took through the agent's reasoning and tool calls can be reconstructed rather than only summarized. Review and optimization surfaces check for on-brand and expected behavior.

Sourcesierra.ai/product/agent-sdkread 2026-09-05

The Agent Data Platform is a named platform layer whose stated job is unifying what the agent knows about a specific customer across sessions, channels and systems. Context persists between conversations rather than being rebuilt per interaction. No scope or retention lifetime is published for that store.

Sourcesierra.ai/product/agent-sdkread 2026-09-05

Sierra is delivered as Sierra-hosted SaaS. Residency is published per subprocessor through the trust portal rather than offered as a region the customer selects, and no VPC, single-tenant or on-premises option is documented. The FedRAMP High commitment implies segregated government infrastructure, but no residency menu the buyer chooses from is stated.

Sourcesierra.ai/product/trust-and-reliabilityread 2026-09-05

The Agent SDK ships with a range of AI skills and systems integrations so customers go live faster, and skills such as triage, respond and confirm are reusable building blocks. These are composable primitives the customer assembles rather than named prebuilt agents, templates or an adoptable catalog, and no library or marketplace is published.

Sourcesierra.ai/product/agent-sdkread 2026-09-05

One agent is built once and deployed across chat, phone, email, SMS and messaging, with the same skills and policies applying on each. Sierra positions a single agent working across every channel as the platform's design point.

Sourcesierra.ai/product/agent-sdkread 2026-09-05

Agents are assembled from more than fifteen frontier, open-weight and proprietary models chosen per job, and Sierra switches between LLM providers automatically to optimize performance and maintain continuity during a provider outage. Sierra states as a selling point that the customer does not hand-pick the models or wire them. The routing happens inside Sierra, and neither customers nor admins choose the model.

Sourcesierra.ai/blog/constellation-of-modelsread 2026-09-05

A published Agent SDK is the developer surface for building, customizing and deploying agents, backed by documentation, interactive training and a development pipeline. Agent OS exposes out-of-the-box tooling for how agents are built, deployed and managed. No public API reference or MCP server for the platform itself is documented, so the developer surface is an SDK rather than an open API contract.

Sourcesierra.ai/product/agent-sdkread 2026-09-05

The Agent SDK verifies that an agent performs as expected across a wide range of scenarios and guards against regressions. Agent definitions are carried through a development pipeline rather than edited live. The artifact under test is the customer's own agent, and the result gates the change.

Sourcesierra.ai/product/agent-sdkread 2026-09-05

Agents act on the customer's systems through SDK-built API integrations held deterministic and policy-bound, and deploy across chat, phone, email, SMS and messaging. No browser control, virtual desktop or operation of software lacking a programmatic interface is documented.

Sourcesierra.ai/product/agent-sdkread 2026-09-05