# Decagon — Enterprise platform for building and running AI customer-support agents across chat, voice and email.

> Source: The Agents Index — https://theagentsindex.com/decagon (structured, researched, re-verified)
> Facts last verified: 2026-09-14

Decagon is an enterprise conversational AI platform for customer support, where teams define agent behavior in natural-language Agent Operating Procedures rather than code. One agent runs across chat, voice and email, with testing, A/B experiments and always-on QA built in.

| Fact | Value |
| --- | --- |
| Website | https://decagon.ai/ |
| Price | We've successfully deployed AI agents across countless customers and support two flexible pricing options |
| Free tier | No |

No free plan, trial or self-serve signup is published anywhere on decagon.ai. /pricing and /plans return 404, the only account link is a sign-in for existing customers, and every call to action is 'Get a demo'; a forward-deployed team accompanies each engagement.

## What it does

**Done-for-you service** · Partly — You run the software and your own CX team owns the AOPs, but Decagon also staffs every engagement with a forward-deployed team, so it is not purely self-service.

> Most enterprises go live in weeks rather than months. A forward-deployed team are included with every engagement to accelerate your team’s path to production.
> — https://decagon.ai/vs/sierra · FAQ, 'How long does Decagon take to deploy?' answer

## How much it does unattended

**Runs autonomously** · Yes — The agent handles inbound customer requests end-to-end across channels without a human, though changes the agent proposes to itself are staged for human approval first.

> Handle inbound requests end-to-end across web live chat, SMS, WhatsApp, and more.
> — https://decagon.ai/product/chat · 'Resolve high-volume inquiries instantly' benefit block

**Multi-agent** · Partly — Several specialised models act on one conversation — a supervisor checking responses and a bad-actor screen — but Decagon never describes customers composing or orchestrating multiple agents themselves.

> during conversations, a supervisor model checks responses in real time and a bad-actor model screens manipulation attempts
> — https://decagon.ai/vs/sierra · FAQ, 'How does Decagon keep agents from going off-script?' answer

**Agent permissions** · Yes — The agent executes real operations such as refunds and identity verification, and how far it may go before escalating is set by AOP guardrails and guidelines.

> Ensure critical operations like identity verification and refunds are executed securely, reliably, and on-brand with strict controls.
> — https://decagon.ai/product/overview · Features section, 'Guardrails and guidelines' card

## Which models it runs on

**Claude** · Partly — Anthropic is named as one of Decagon's AI providers and is listed as an LLM subprocessor, but the buyer never selects Claude — routing is Decagon's.

> Decagon enforces zero-day retention with all AI providers like OpenAI and Anthropic, ensuring no conversation data is stored or used for training.
> — https://decagon.ai/security · 'Protect your data' section, 'Zero-day retention with LLMs' card

**GPT** · Partly — OpenAI is named as an AI provider and appears as an LLM subprocessor on the trust centre, yet no page offers the customer a choice of GPT.

> Decagon enforces zero-day retention with all AI providers like OpenAI and Anthropic, ensuring no conversation data is stored or used for training.
> — https://decagon.ai/security · 'Protect your data' section, 'Zero-day retention with LLMs' card

**Model choice** · No — Model selection belongs to Decagon, not the buyer: the engine is proprietary and most traffic runs on models Decagon trained itself, with no published model picker.

> Today, over 80% of all model traffic at Decagon runs on models we trained ourselves.
> — https://decagon.ai/blog/introducing-decagon-labs · Opening paragraph of 'Introducing Decagon Labs'

_Not established: Open models_

## Where you use it

**In your pipeline** · Partly — Agent versions can be tested from CI/CD pipelines and versioned in Git, but nothing describes Decagon acting on a pull request itself.

> Integrate with CI/CD practices to test every agent version and ensure every change meets the quality bar without regressions.
> — https://decagon.ai/product/testing-qa · 'Ensure high performance at enterprise scale' section, 'Ship every update with confidence' card

**In a browser** · Yes — The platform is a hosted web console: building, monitoring, QA dashboards and analytics all run in-product behind a sign-in.

> Track trends across flags and categories to spot systemic issues, monitor progress, and prove impact from a single source of truth shared across teams.
> — https://decagon.ai/product/watchtower · 'Turn support data into customer intelligence' section, 'Real-time dashboards' card

_Not established: On the command line_

## Whose machine it runs on

**Self-hosted** · Partly — Decagon claims headless deployment inside your own infrastructure, but publishes no on-premise or self-hosted edition and runs its platform on its own cloud subprocessors.

> tools are written in Python with MCP support, and the platform deploys headlessly inside your existing infrastructure
> — https://decagon.ai/vs/sierra · FAQ, 'How does Decagon fit into developer workflows?' answer

**Open source** · No — No repository, licence or source availability appears anywhere on the site; the vendor markets the engine underneath as proprietary.

> Every Decagon agent is built on a proprietary engine that understands your business and becomes more intelligent with every interaction.
> — https://decagon.ai/product/overview · 'Agent engine' section, intro paragraph

## What it costs to run

**How it meters** · Yes — Billing meters customer contact, not seats: a fixed rate per incoming conversation, or a higher rate per resolution with escalations free.

> Per-conversation pricing: A fixed rate for every incoming conversation, with flexible pricing for higher volumes.
> — https://decagon.ai/blog/pricing-ai-agents · 'The Customer Experience Use Case' section, pricing options list

**Free tier** · No — Every entry path on the site is a sales demo: there is no signup, no trial and no free plan, and each engagement ships with a forward-deployed team.

> Most enterprises go live in weeks rather than months. A forward-deployed team are included with every engagement to accelerate your team’s path to production.
> — https://decagon.ai/vs/sierra · FAQ, 'How long does Decagon take to deploy?' answer

**API access** · Partly — Agent logic and tools are Git-backed and written in Python and the platform deploys headlessly, but the API reference sits behind a login so the public surface cannot be checked.

> Decagon is built to feel native to an engineering team: agent logic and tools are Git-backed with versioned workspaces for staging and production
> — https://decagon.ai/vs/sierra · FAQ, 'How does Decagon fit into developer workflows?' answer

**MCP server** · Partly — MCP is supported as the way Decagon's agent reaches your systems; no page says Decagon exposes an MCP server another agent could query.

> Connect to any data system or application via an open standard for real-time data and actions.
> — https://decagon.ai/product/integrations · 'Extend your support stack' section, 'MCP open connectivity' card

_Not established: Bring your own key_

## Buying it for a team

**A company can buy it** · Yes — This is sold to enterprises only — named logos are large organisations and the buying motion runs through a demo and a signed engagement.

> Our platform enables leading enterprises like Avis Budget Group, Chime, Oura Health, 1-800-FLOWERS.COM, and Hunter Douglas to deploy AI agents
> — https://decagon.ai/about · 'About Decagon' section, second paragraph

**Seat model** · No — Pricing is metered on conversations or resolutions rather than per seat, and no seat floor, ceiling or per-user rate is published anywhere.

> Sierra prices per outcome (per resolution), while Decagon lets you choose per-conversation or per-resolution pricing.
> — https://decagon.ai/vs/sierra · FAQ, 'How does Decagon's pricing compare to Sierra's?' answer

**Admin controls** · Yes — Admins constrain the agent two ways: role-based platform permissions, and AOP guardrails that gate refunds, escalations and brand voice.

> Platform access is defined and enforced by role, ensuring every user automatically gets the right permissions for their job.
> — https://decagon.ai/security · 'Authenticate users' section, 'Role-Based Access Control (RBAC)' card

**Audit log** · Partly — Tamper-protected audit logs exist and capture logins and unusual activity, but Decagon states access is limited to its own senior engineering leadership rather than the customer's admins.

> Comprehensive, tamper-protected logs capture key events, from logins to unusual activity, accessible only to senior engineering leadership.
> — https://decagon.ai/security · 'Ensure safe, transparent, and accountable AI performance' section, 'Audit logs' card

**Single sign-on** · Yes — SSO is offered against enterprise identity providers, with Okta and Microsoft Entra named; the underlying SAML or OIDC protocol is not stated.

> Decagon integrates with identity providers like Okta and Microsoft Entra, enabling secure access across systems without additional passwords.
> — https://decagon.ai/security · 'Authenticate users' section, 'Single Sign-On (SSO)' card

_Not established: Pooled budget_

## What happens to your code

**Opt out of training** · Yes — Stronger than an opt-out: zero-day retention is enforced with every AI provider by default, so conversation data is not stored or trained on.

> Decagon enforces zero-day retention with all AI providers like OpenAI and Anthropic, ensuring no conversation data is stored or used for training.
> — https://decagon.ai/security · 'Protect your data' section, 'Zero-day retention with LLMs' card

**Data residency** · Partly — Default processing is US-hosted on Google Cloud; EU-only residency inside EEA infrastructure is available on request.

> Decagon offers EU-only data residency, in which Customer Personal Data is processed and stored within EEA-located infrastructure.
> — https://decagon.ai/legal/dpa-security-annex · Annex 4 - Security Measures, 'Measures for cross-border data transfers and data residency'

**Getting out** · Partly — Decagon asserts full data portability and no SDK lock-in, and its trust centre lists deletion on departure, but no export API or retention terms are published.

> Complement your systems of record with full data portability, while retaining granular control over what context is stored and how it’s used.
> — https://decagon.ai/proactive-agents · 'User Memory' section, 'Enterprise-grade governance' card

**Certifications** · Yes — A full enterprise set is claimed and the trust centre carries the underlying reports, including a SOC 2 Type II report and an ISO 27001 certificate.

> SOC 2 Type II, HIPAA, GDPR, PCI, and ISO 27001
> — https://decagon.ai/vs/sierra · 'How Decagon and Sierra compare' table, 'Compliance certifications' row, Decagon column

## Provenance

Every fact above comes from the vendor. Nothing here is independently corroborated yet.

## Coverage elsewhere

Pages about this product that the vendor does not control. The label states how independent each one is.

| Source | Kind | Read |
| --- | --- | --- |
| en.wikipedia.org | Platform record | 2026-08-23 |
| siliconangle.com | Independent source | 2026-08-23 |
| bloomberg.com | Independent source | 2026-08-23 |
| fin.ai | A competitor's comparison | 2026-08-23 |
| eesel.ai | Independent source | 2026-08-23 |
| quiq.com | Independent source | 2026-08-23 |
| g2.com | Review platform | 2026-08-23 |

## What changed

- 2026-09-14 — Capabilities · Commercial terms · Free tier · Price
- 2026-09-02 — Capabilities · Commercial terms · Free tier
- 2026-08-31 — Capabilities · Commercial terms · Free tier
- 2026-08-22 — Alternatives · One-liner · Summary

## Correcting this record

Anything marked "Not established" is a question we have not answered — it is not a No. If you represent Decagon, or anything here is wrong or out of date, you can supply it with a source: the operations this site accepts from an agent are listed at https://theagentsindex.com/.well-known/agents.json — that catalogue needs no key to read, and each entry states its own auth.

Supplying evidence is free. Nothing that can be bought here changes what a field says, whether a gap is closed, or how this record is scored.