Sierra
Type: app · Vendor: Sierra · Language: Web product · License: proprietary · Status: active · Status in practice: mature
Sierra Agent OS: build one production-grade CX agent from skills (triage / respond / confirm), goals and guardrails, deploy across chat, voice, SMS, WhatsApp, email and ChatGPT, with memory across conversations and an LLM-as-judge monitoring loop.
Description. Sierra sells an Agent OS plus an Agent SDK that productises building a single agent and deploying it across many channels. The SDK composes skills (triage, respond, confirm) into multi-step workflows under explicit goals and guardrails, integrates customer knowledge, takes secure actions, and routes escalations to humans. Agent OS 2.0 (Nov 2025) added memory and an Agent Data Platform so agents can connect dots across time. Ghostwriter generates a production-ready multilingual multichannel agent with built-in guardrails. A blog series describes context engineering and LLM-as-judge monitoring as core operational tooling.
Agent loop shape. Agent SDK composes skills (triage/respond/confirm/...) into a workflow under explicit goals and guardrails. Per-customer context (history, real-time signals) feeds personalised responses; the agent can take secure actions and escalates to humans when needed. Agent OS 2.0 adds memory and an Agent Data Platform so the same agent persists context across conversations. Voice runs on a transcription platform supporting 70+ languages; quality is monitored with LLM-as-judge.
Primary use cases
- single CX agent across chat, voice, SMS, WhatsApp, email and ChatGPT
- skill-composed multi-step workflows with goals and guardrails
- voice-first CX with real-time transcription and benchmarks
- agent monitoring with LLM-as-judge
Key concepts
- Agent SDK (docs) — Goal-directed builder that composes skills into multi-step workflows.
- Skills → agent-skills (docs) — Composable building blocks like triage, respond and confirm.
- Guardrails (docs) — Explicit objectives and policy constraints on agent behaviour.
- Agent OS 2.0 + Agent Data Platform → cross-session-memory (docs) — Memory and context layer letting the agent connect dots across time.
- Ghostwriter (docs) — Generates a production-ready multilingual multichannel agent.
- LLM-as-judge monitoring → llm-as-judge (docs) — Quality evaluation loop for live agents.
Patterns this app implements —
- ★★Agent Resumption
Agent OS 2.0 + Agent Data Platform give agents memory and continuity across conversations.
- ★★Conversation Handoff to Human
Vendor documents intelligent routing to the right team member on escalation; Live Assist gives the human handler the AI agent's context.
- ★★Tool Use
Agents take secure actions to support customer requests; tool calls and knowledge lookups are observable.
- ★Agent Skills
Agent SDK composes named skills (triage, respond, confirm) into goal-directed multi-step workflows.
- ★★Input/Output Guardrails
Goals + guardrails are explicit primitives in the SDK; Ghostwriter ships agents with built-in guardrails.
- ★Multilingual Voice Agent Stack
Vendor publishes a transcription platform with 70+ languages and a real-time voice agent benchmark (τ-voice).
- ★★LLM-as-Judge
Vendor blog explicitly describes LLM-as-judge as the monitoring loop for production agents; an independent judge grades whether the agent achieved the user's goal.
- ★★Approval Queue
Vendor documents an Agent Assembly Line shape where Ghostwriter validates proposed changes in a sandboxed environment and prepares them for human review before shipping; not a per-action runtime HITL…
- ★★Session Isolation
Vendor describes per-customer context personalisation drawn from conversation history; ADP unifies what the company knows about a customer, but no published per-process runtime isolation primitive. H…
Neighbourhood
Click any neighbour to follow the lineage. Scroll to zoom, drag to pan.
Alternatives & relatives
Listed as alternative by (4)
References
Provenance
- Last analyzed:
- Last updated:
- Verification status: verified