Decagon

What it does

Decagon is a conversational AI platform for customer support. Teams define agent behavior with Agent Operating Procedures written in natural language, and the agents handle customer conversations across voice, chat, and email. The agents connect to a company's existing support tools so they can pull live data and trigger workflows.

People use Decagon to build, test, and improve support agents without relying on a lengthy engineering cycle. Duet analyzes conversation transcripts to suggest and draft workflow changes, while technical teams can version agents with Git-based tracking, run simulated conversations, and A/B test updates. Its distinct emphasis is traceability, showing the model calls, workflow triggers, and knowledge articles behind each decision.

Decagon suits customer experience and operations teams managing support across several channels. Business staff can describe procedures, while technical teams control integrations, versioning and operational safeguards. Access begins with a demo request to discuss the intended workflows and systems. Before launch, teams should test the agent against realistic exceptions, check the actions it can perform and define handoff to human support. Conversation traces and review tools help diagnose behavior and guide later changes.

What it can do

• Build customer support workflows in natural language using Agent Operating Procedures
• Deploy agents across voice, chat, and email from a single intelligence layer
• Integrate with ticketing platforms, CRMs, knowledge bases, and CCaaS providers
• Test agents with simulated conversations and unit testing before rollout
• Version agents and run A/B tests against real conversations
• Trace model calls, workflow triggers, and referenced knowledge articles for each decision

Example workflows

Building an inbound support workflow

Illustrative example. Input: a support team's standard procedure for handling refund requests. Steps: write the procedure in natural language as an AOP, add guardrails for identity verification, and test it with simulated conversations. Output: an agent workflow that runs on live chat, with each step traceable for review.

Improving an agent from transcripts

Illustrative example. Input: uploaded customer conversation transcripts. Steps: Duet analyzes the conversations to identify workflow gaps and drafts AOP revisions, then the team tests the revisions and A/B tests them against real conversations. Output: a versioned agent update with recorded test results.

Before you choose

• Review agent procedures, data permissions and escalation paths before launch.
• Plan for integration work and version control alongside business-user configuration.
• Confirm commercial, support and deployment terms during the demo.

Listed under

• Support
• Chat and email agents
• Voice and phone agents
• AI agents
• Chat
• Conversational AI
• Customer experience
• Customer service agents
• Customer support

Get started

The site directs visitors to a demo request to discuss their customer experience needs.

1. Define the first agent workflow as an Agent Operating Procedure in natural language.
2. Connect the agent to the existing ticketing, CRM, or knowledge base tools it needs.
3. Set guardrails for escalations and sensitive operations, then test with simulated conversations before deployment.

Request a demo and confirm pricing, usage scope, onboarding and support for your deployment.

Decagon doesn’t publish prices. You get a quote by contacting sales. See their pricing page ↗

Sources for this overview

Product pages, documentation, and pricing checked 29 Oct 2026. Each source supports part of the overview; availability and prices can change.

• Products for Conversational AI | Decagon ↗decagon.ai
• Get a Demo | Decagon ↗decagon.ai
• Agent Operating Procedures (AOPs) | Decagon ↗decagon.ai

Company website

decagon.ai: product, pricing and policy pages

YC directory mirror

Batch and company name

Social accounts

Accounts the company links, and its posts on X

Human review

Checked by a person

knowledge_base

Last checked 29 Oct 2026. Every field keeps its source and date; unknown means we couldn’t confirm it.