Duckie
Duckie is an agentic AI platform for customer support teams, built to let non-engineers ship working agents. Its agents connect to your helpdesk, knowledge base and business tools, then resolve tickets end to end, investigate backend issues and automate the workflows behind support, with replay testing, policy guardrails and a full trace of every run before anything reaches a customer.
What is Duckie?
Duckie is an agentic AI platform for customer support teams, built by Duckie AI, aimed at companies that want agents to close tickets rather than suggest canned replies. The premise is simple: a support lead should be able to describe an outcome in plain English, such as cancelling a subscription in Stripe, filing a Jira ticket for a suspected data bug or chasing a stalled payment, and Duckie turns that description into an agent that plans, calls tools, verifies the result and replies to the customer.
Under the hood the product is a control plane rather than a single chatbot. Agents are assembled from a knowledge base synced out of Notion, Confluence, Google Drive, Guru or Skilljar, or written directly inside Duckie; from runbooks that describe procedures step by step; from guidelines that set tone; from guardrails such as escalation rules and hard restrictions; and from tools drawn from more than twenty integrations, including Zendesk, Intercom, HubSpot, Freshdesk, Plain, Pylon, Slack, Discord, Gmail, Jira, Linear, GitHub and a website widget, plus custom HTTP tools and MCP servers. A visual canvas orchestrates several agents at once, with triage, investigation, resolution and escalation agents handing work to each other, and model choice is set per agent and switchable among frontier providers.
Production readiness is where Duckie separates itself from demo-grade bots. Workflows can run in testing, shadow or live mode, and replay testing re-runs historical tickets to compare agent output against what human agents actually did. A policy engine enforces rules across every agent: redacting PII before tool calls, requiring human approval above a spend threshold, blocking sends to external domains and scoping permissions. Every run keeps a full trace of reasoning, retrieved knowledge, tool calls and retries, and a kill switch can halt an agent instantly.
Deployment covers helpdesk tickets, Slack and Discord channels, Gmail and an embeddable website widget, with webhooks and scheduled runs for work that customers do not initiate. Analytics track resolution rules, categories, attributes, run history and knowledge gaps.
Duckie publishes customer results rather than promises: 90% ticket automation at Grid, 95% resolution at Vanquish Trader, 92% deflection at Automox and a 66% support cost reduction at Abbiamo. It does not publish a price list; plans are discussed on a demo call.
The company is small and engineering-led. Co-founders Valerie, formerly a senior engineer at LinkedIn who built the first version to escape repetitive on-call questions, and Joel, formerly at Netflix and C3 AI, frame the product around reliability rather than conversational flair. Duckie publishes a status page, carries a SOC 2 Type II badge and documents which model providers and subprocessors touch customer data.
Duckie Core Features
Plain-English agent building
Describe the outcome you want, attach tools and Duckie compiles it into an agent with planning, tool-call and verification steps, no engineering required.
Ticket resolution end to end
Agents read the ticket, look up the account in connected systems, take the action, verify it landed and reply, then close the ticket.
Multi-agent orchestration canvas
Triage, investigation, resolution and escalation agents pass work to each other on one visual surface instead of running as isolated bots.
Replay and shadow testing
Re-run thousands of real historical tickets, compare agent output with human handling, and run agents in shadow mode before they touch a customer.
Policy engine and guardrails
PII redaction before tool calls, approval thresholds above a spend limit, blocked external domains, permission scoping and an armed kill switch.
Full run traces
Every run records reasoning, knowledge retrieved, tool calls, retries and outcome, so support leads can debug and improve agents without engineering help.
20+ native integrations
Zendesk, Intercom, HubSpot, Freshdesk, Plain, Pylon, Slack, Discord, Gmail, Jira, Linear, GitHub, Notion, Confluence, Google Drive, Sentry and more, plus custom HTTP tools and MCP servers.
Model flexibility and self-hosting
Switch frontier models per agent, restrict to AWS Bedrock models, or deploy the whole platform on your own infrastructure for data sovereignty.
Who is Duckie for?
Duckie is built for the people who own customer support operations, not for developers looking for a chat widget SDK. It fits best in these situations: Support leaders at B2B SaaS companies - heads of support and CX who are measured on resolution time and cost per ticket, and who want AI to close routine and semi-technical tickets rather than deflect them to a help article. E-commerce and subscription support teams - businesses handling cancellations, refunds, plan changes and failed payments, where an agent that can read a subscription in Stripe and act on it saves more time than one that only answers questions. Technical support and escalation teams - companies whose tickets involve backend investigation, log reading or bug filing, where Duckie's tool calls into Sentry, Jira, Linear and GitHub matter more than conversational polish. Small teams carrying large volumes - startups with one to five support staff handling thousands of tickets a month, the profile behind Duckie's published Abbiamo and Vanquish Trader case studies. CX and RevOps leads consolidating point tools - teams already running Zendesk, Intercom, HubSpot, Freshdesk, Plain or Pylon who want one control plane over several agents instead of a separate bot per channel. Security-conscious enterprises - organizations that need self-hosting, AWS Bedrock-only models, PII redaction, permission scoping and audit trails before letting an agent touch production systems. It is a weaker fit for solo founders who want a five-minute website chatbot, for teams with no documentation or clean helpdesk history to learn from, and for anyone who needs a published price and a credit card checkout before they can start. Teams usually arrive because a single agent builder cannot cover both the conversational and the operational side of support. Answering a question is easy; checking an order, changing a plan, chasing a failed payment or filing a bug with the right context is where the queue actually sits. That gap is what Duckie targets, which is why its strongest references come from fintech, developer tools and subscription businesses rather than from simple FAQ-heavy sites.
Duckie Use Cases
Resolve subscription and billing tickets that require reading Stripe and updating a plan
Deflect repetitive how-to questions by answering from your synced knowledge base
Investigate suspected bugs by pulling logs from Sentry and filing a Jira ticket with context
Triage incoming tickets, tag them by category and route escalations to the right team
Handle refund and cancellation requests with policy checks and human approval above a threshold
Draft and send replies in Slack, Discord or email without leaving the support channel
Replay last quarter's tickets to prove an agent's resolution rate before going live
Run scheduled support operations work such as daily digests or stalled-payment outreach
Duckie Pros and Cons
Pros
- Agents take real actions in connected systems rather than only drafting replies, which is what actually removes tickets from the queue
- Testing and safety tooling is unusually complete for this category: replay testing, shadow mode, policy engine, PII redaction, permission scoping and a kill switch
- Published customer outcomes are specific and varied, from 90% ticket automation at Grid to 66% support cost reduction at Abbiamo
- Broad integration surface with 20+ native connectors plus MCP and custom HTTP tools, so unusual stacks are not blocked
- Self-hosting, AWS Bedrock-only models and SOC 2 Type II posture make enterprise security review more tractable
Cons
- No published pricing anywhere on the site, so evaluation starts with a sales call instead of a self-serve trial with known limits
- Quality depends on clean documentation and helpdesk history, which many teams discover only after connecting a messy knowledge base
- Advanced setup such as custom HTTP tools, MCP servers and self-hosting still needs technical staff despite the no-engineer positioning
FAQ About Duckie
Duckie Pricing
Duckie does not publish pricing. The /pricing URL returns a 404 and every commercial path on the site leads to a demo booking, so plan tiers, message or ticket volumes, self-hosting and support terms are quoted on a call. You can create an account and book a demo without a card, but there is no advertised self-serve plan with published limits, and third-party figures circulating online are unconfirmed by the company.
Check official pricingCustom
Pricing discussed on a call; plan tiers, volumes and self-hosting terms are quoted after a demo.
Duckie Alternatives
Motion
Motion is a video agent from Mosaic that turns a written brief, raw footage or a reference clip into an edited video. It plans the story, generates or cuts shots, adds music and sound, then lets you refine the result in chat, on the preview or in a timeline before exporting to Premiere Pro, After Effects or DaVinci Resolve.
Cotlet
Cotlet is a hosted MCP server and REST API for apps with no API. Cotlet builds and maintains each integration and fixes it when the app changes. Cotlet is operated by Take Group sp. z o.o., Katowice, Poland.
Buddy
Buddy is a DevOps and platform-engineering automation suite for building, testing and shipping software. It pairs visual and YAML CI/CD pipelines with deployments to thousands of targets, ephemeral cloud sandboxes, tunnels, domains and visual regression tests. Its AI angle is agentic: a hosted MCP server, the bdy CLI and agent skills let Claude Code, Codex and Cursor run pipelines and deploy.