Last updated: 2026-10-19
Duckie, from the Y Combinator Winter 2024 batch, lets support teams build AI agents that answer and act inside Zendesk, Intercom, Slack and a long list of other connected tools. Agents follow runbooks and guardrails, are tested against past tickets, and the platform can be self-hosted on Docker Compose, Kubernetes or AWS.
About Duckie
Duckie is a platform that lets a support team build and run its own AI agents instead of buying a fixed chatbot. It began as an on-call helper that co-founder Valerie wrote to stop answering the same questions, then spread to other teams and other companies. The company went through Y Combinator in the Winter 2024 batch and is based in San Francisco.
The working model is plain: you describe how a ticket should be handled in ordinary English, point the agent at your documentation, and decide which actions it may take. A refund request, a plan change or a bug report becomes a procedure the agent follows and a human can read afterwards. That makes it closer to a workflow builder for support leads than to the answer-bot approach of Intercom or Zendesk add-ons.
The buyer is a support or operations lead who has engineers in short supply. Enterprise-managed agents such as Decagon and Sierra put the vendor's implementation team between you and the agent; Duckie pushes that work to your own team and gives them testing tools to do it safely. Parahelp and eesel sit nearer to it on the build-it-yourself side. Lighter chat-first options include Ada and Chatbase. Kapa.ai fits when the job is answering product-docs questions, and Vapi is the one to look at for phone and voice support, which Duckie does not list. Browse the full customer support category for the wider field.
Pricing is not published. The site sends visitors to a demo booking, so the first real number comes from a sales call. It connects to HubSpot, Pylon and Notion, and a GitHub connection lets it read repositories and issues. Under the hood the vendor lists Anthropic and OpenAI as model providers and AWS for hosting, and it offers a self-hosted build for teams that cannot send tickets to someone else's cloud. If you are unsure whether it fits, try Smart Match with your ticket mix and tooling.
Pricing
Duckie does not publish pricing. The pricing page returns a 404 and the vendor site routes buyers to a demo booking. There is no free tier or self-serve checkout listed.
Get a quote and ask whether it is charged per seat, per run or per resolution, and whether the self-hosted build is priced differently. For a usage-billed contrast, see eesel.
Key Features
- Runbooks and sub-agents: Agents follow written runbooks and can hand parts of a ticket to sub-agents, laid out on one orchestration canvas.
- Actions inside your stack: Agents can act in Zendesk, Intercom, HubSpot, Freshdesk, Plain, Pylon, Slack, Discord, Gmail, Jira, Linear and GitHub, 20+ integrations on the vendor's list.
- Knowledge from existing docs: Reads Confluence, Notion, Google Drive, Sentry, Fireflies, Guru and Skilljar, so answers come from the material your team already keeps.
- Guidelines and guardrails: Tone rules and safety limits are set per agent, and conversations that need a person are escalated to one.
- Replay, batch and playground testing: An agent can be run against past tickets, in a batch or in a sandbox before it goes near live customers.
- Run history and analytics: Every run keeps its reasoning, the knowledge it used and the actions it took, with alerts and a breakdown of what customers ask.
- Self-hosting: Runs on Docker Compose, Kubernetes or AWS Terraform, with Qdrant for vectors, Redis for queues and Supabase or Postgres for data.
- Switchable models: The vendor says teams can switch between frontier models whenever they want, and lists Anthropic and OpenAI as providers.
Pros
- A self-hosted build keeps ticket data inside your own cloud account, an option that hosted-only support agents cannot offer by design.
- Support leads can write and test agents in plain English, which suits teams without spare engineering time.
- Case studies on the vendor site span fintech, logistics, healthcare technology and endpoint management, so it is not tied to one vertical.
Cons
- No public pricing, and every plan starts with a demo call, so budgeting before a sales conversation is guesswork.
- Small company: its YC profile lists a team of 4, so confirm support coverage and roadmap commitments in the contract.
- SOC 2 status is unclear: the footer badge says Type II but the January 2026 security text says the work was in progress, so ask for the report.
- Headline results such as 90% ticket automation are the vendor's own case-study claims, not independent benchmarks.
Data Handling
- Training-data policy
- Security page says customers keep all rights in their data and Duckie reserves no right to use or disclose it.
- Compliance
- SOC 2 Type II (vendor badge)
Frequently Asked Questions
How much do you pay for Duckie?
Duckie does not publish prices. Its pricing page returns a 404 and the site sends buyers to a demo booking, so a quote is the only way to get a figure. When you ask, find out what the unit of charge is and whether the self-hosted build costs the same as the cloud one.
Is Duckie SOC 2 compliant?
The site footer shows a SOC 2 Type II Certified badge. The security page, last updated January 2026, still says the company was working toward SOC 2 with a compliance-automation partner. Ask for the audit report and its period before you sign; the security page says customers receive independent audit reports once finalized.
What should you use instead of Duckie?
Decagon suits enterprise teams that want a vendor-run rollout across chat, email and voice. Sierra is the other enterprise option, with outcome-based pricing. Parahelp is a similar quote-only support agent, and eesel is a lighter, usage-billed choice if you want to start without a sales call.
Is Duckie better than Decagon?
They answer different buyers. Decagon is aimed at large enterprises that want the vendor to run implementation, while Duckie hands the building to your own support team and offers a self-hosted build. Neither publishes a price, so compare written quotes and ask each for reference customers on your helpdesk.
How long does it take to get going with Duckie?
The quickstart in the docs says a first agent takes about 10 minutes, and the vendor says production deployment takes around two weeks. Plan for more: you will write runbooks, connect a helpdesk such as Zendesk or Intercom, and run replay tests before letting an agent answer real customers.
Top Alternatives
- Decagon: Pick Decagon if you want a vendor-run enterprise rollout; pick Duckie if your own support team will build and own the agents.
- Sierra: Pick Sierra for outcome-priced enterprise agents with vendor implementation; pick Duckie if you want to configure agents in-house.
- Parahelp: Pick Parahelp for a quote-only support agent tuned to ticket resolution; pick Duckie if you need runbooks, sub-agents and a self-hosted build.
- eesel: Pick eesel if you want usage-billed ticket automation you can start without a call; pick Duckie if you need deeper custom actions.