Overview
Publish any Taskade AI agent to a public URL so customers, prospects, or community members can chat with it directly. You decide which tools the agent can call when used by anonymous visitors and which projects feed its context. Everything else stays private inside your workspace.
TL;DR: Publish a custom AI agent at
taskade.com/a/{agent-name}-{id}. Choose its appearance, access settings, starter prompts, and knowledge permissions. Visitors can chat without entering your workspace. Their conversations appear in the Agent Inbox. You can also embed the agent on a website.
Publish an agent
Open the agent in the Agents tab, then turn on public access.
- Open the Agents tab in your workspace or folder.
- Click the agent you want to publish.
- Open Share in the top-right.
- On the Share tab, toggle Public access on.
- Click Save, then copy the public link.
The agent is reachable immediately. The link carries the agent's name followed by its id, so an agent called Support Bot gets something like taskade.com/a/support-bot-01JK82XFM2. Rename the agent and the readable part changes, but the old link keeps working because the id at the end is what Taskade looks up. Send the link, drop it into a Notion page, or post it in your community.
The publish panel keeps every control in one place, split across Share, Embed, Customize, and, when the agent has tools, Tool Permissions. A live preview on the right shows the agent exactly as visitors will see it while you set things up.
┌────────────────────────────────────────────────────────────────┐
│ Publish Agent │
│ [ Share ] Embed Customize Tool Permissions │
├────────────────────────────────────────────────────────────────┤
│ Public access [ On ● ] │
│ Public link [ taskade.com/a/support-bot-01JK… ] Copy │
│ │
│ Password protect [ Off ○ ] │
│ Password [ •••••••••••• ] │
│ │
│ Copy Knowledge [ On ● ] │
│ │
│ Preview → full page | Chat Widget │
└────────────────────────────────────────────────────────────────┘
What each publish control does
Use this table to decide which controls to turn on before you share the link.
| Control | What it does |
|---|---|
| Public link | Gives the agent a clean URL at taskade.com/a/{agent-name}-{id} that anyone can open. |
| Password protection | Locks the link so only people with the password can use it. Best for client-only or internal-only agents. |
| Embed widget | Drops the agent onto your own website as a chat launcher. See Agent Widgets. |
| Copy Knowledge | Lets people copy the agent's knowledge when they clone it. |
Customize the public surface once
The Publish Agent panel splits into Share, Embed, and Customize, plus Tool Permissions when the agent has tools. Set the look on Customize and every place the chat appears picks it up. You are describing the agent, not decorating one particular page.
- Theme: Light theme, Dark theme, or Auto.
- Color: pick a hex code for the launcher and accents. Taskade works out whether light or dark text reads better on it, so your label stays legible.
- Header: keep Default, which uses the agent's name, or switch to Custom and write your own title.
- Agent avatar: the picture shown beside the agent's replies.
- Chat button icon: an emoji or an uploaded square image on the floating chat button. Leave it unset and the agent's avatar is used.
- Message Style: Default for a transcript-style thread, or Bubble for rounded chat bubbles.
- Show Suggestions: turn the starter-prompt chips on or off. The prompts themselves come from the agent's conversation starters.
- Notice: one short dismissable line above the chat, for a privacy notice or opening hours. Up to 200 characters, and a visitor's dismissal lasts for that browser session.
- Footer: one short plain-text line under the composer, for a disclaimer or a legal note. Up to 200 characters.
- Background: an image behind the chat.
- Hide Branding: removes the Taskade mark from the public chat.
A live preview sits beside the panel, and you can flip it between the full page and the Chat Widget form factor. Changes save instantly, and the public surface picks them up on the visitor's next load.
Know where those settings show up
Three surfaces read the same appearance settings, so there is nothing to configure twice.
The website widget is the public page in a floating panel, which is why the two always match. A Taskade Genesis app's chat block reads the same profile, so the assistant inside your app shows your agent's name, your starter prompts, your placeholder text, and your accent color rather than a generic "Assistant".
The widget snippet carries the agent's saved theme and accent color. Per-site settings cover the launcher corner, size, label, welcome bubble, and auto-open delay, so the same agent can sit differently on two websites without changing its public-page appearance.
Check which appearance settings need a plan
Most of the appearance controls work on every plan. Three do not.
| Setting | Plan |
|---|---|
| Theme, Color, Header, avatar, chat button icon, Message Style, Show Suggestions, Notice, Footer | Every plan |
| Hide Branding | Business and up |
| Background | Business and up |
| Preview image, the link-preview picture for the public page | Business and up |
Those three carry a Business badge in the panel, so you can see the plan requirement before you commit to a design. Turning one of them back off, or removing an image you already uploaded, is never blocked, so a workspace that changes plan can always undo a setting it can no longer switch on. Current plan details are on the pricing page.
Scope public agent knowledge
Public agents only see the knowledge sources you have explicitly attached to the agent. The Knowledge panel is the source of truth: anything connected there is fair game for public answers. Anything outside the agent stays private. Connect a public FAQ project, leave your internal customer database (CRM) out, and the agent answers from the right material every time.
For sensitive content, build a separate public-facing agent with its own knowledge connections instead of toggling sources on and off in real time. That keeps the security model simple. See Agent Knowledge & Memory for the full setup.
Switch Between Search and Chat
Published agents offer two ways to use the same approved knowledge.
- Search finds a focused answer from the agent's public knowledge.
- Chat continues a conversation with follow-up questions and tools that you allowed for visitors.
Visitors can switch modes without leaving the public agent page.
How it fits Workspace DNA
A public agent is the public-facing tip of your Workspace DNA. It draws Memory from the Projects you connect, runs Intelligence with the agent's prompt and tools, and triggers Execution through whichever automations it can call.
Anything you do not connect to the agent stays internal. That is the security model in one picture.
Public Inbox
Every public conversation lands in the agent's Inbox. Read it like a chat support log, reply, or hand off to a teammate.
- Open the agent.
- Click Chat in the top-right.
- Switch to the Visitor Inbox tab in the left sidebar.
You will see every public thread grouped by visitor, with the original prompt and the agent's full response. Reply manually, or wire up an automation to forward unresolved threads to Slack, email, or a CRM ticket. Read Agent Inbox for the full setup.
Rate limits and abuse controls
Public agents come with safeguards by default. A rate limit is a cap on how many messages one visitor can send in a short window, and your agent gets one automatically.
- Free conversations are capped per visitor, so a single user cannot drain your AI credits in one session.
- Signed-in conversations consume the workspace credit pool, governed by your plan.
- Every public agent automatically slows down rapid-fire traffic from one internet address, which stops most bots.
- You can pause public access in one click. The link stops working instantly and shows a "page not found", and you can resume any time.
The cap runs as two budgets. One covers the whole agent, and a second covers each visitor, so one abuser can spend only a fraction of the agent's budget before the rest of your visitors are affected. Messages that fail those checks are never charged against the agent's budget, so a bot cannot drain it with junk traffic. Plan-level credit caps apply at the workspace level, so a viral public agent never silently overdraws.
Conversation transcripts stay private to the workspace by default. A member can explicitly create a revocable read-only link for one conversation. Anyone with that link can read that shared transcript until a member disables the link. Other inbox conversations remain private.
Lock or take down an agent
Add a password from the Share tab when you want a published agent reachable by link but limited to the right people. Visitors enter the password before they can chat, so a client-only or internal-only agent stays out of public hands while still living at a clean URL.
Switch the Public access toggle off and the public URL returns 404 immediately. Knowledge connections, tool configurations, and conversation history all stay intact, so you can flip the agent back on later without losing setup.
This is useful for:
- Limited-time campaigns. Publish for a launch week, then take it down.
- Iterating in private. Pause, retrain, re-enable.
- Compliance reviews. Take an agent offline, audit, and restore.
Related guides
- Embed AI Agents as Website Widgets: Drop a public agent onto your site as a chat launcher.
- Public Agent Inbox: Read and reply to every public conversation.
- Tools for AI Agents: Pick which capabilities a public agent can call.
- Custom AI Agents: Build the agent before you publish it.
- Agent Knowledge & Memory: Train an agent on the docs you want it to share.