TL;DR: A Custom AI Agent in Taskade is a configurable teammate built from a persona, knowledge sources, 22+ built-in tools, and a frontier model. Agents are the Intelligence layer of Workspace DNA — they read from your Projects (Memory) and trigger Automations (Execution). Create your first agent free →
What Is a Custom AI Agent?
A Custom AI Agent in Taskade is a reusable, configurable AI teammate that combines a persona, knowledge sources, tools, and a frontier model into a single addressable assistant your team can chat with, embed, or call from automations. Unlike a one-off chat session, an agent has persistent memory across conversations, can act on your workspace, and can be shared publicly with a single link.
Agents are how Taskade turns raw AI capability into something that does work — research, support, qualification, code review, content drafting — instead of just answering questions.
The Intelligence Layer of Workspace DNA
Agents are the Intelligence pillar that sits between Memory and Execution. Projects store what your team knows; Automations execute what your team decides; Agents are the brains that read context, make decisions, and pull the trigger.
The loop is self-reinforcing: every action an agent takes creates new Memory, which makes the next decision smarter.
Anatomy of an Agent
Every custom agent has six configurable parts. Get the right combination and the agent feels like a real teammate instead of a chatbot.
| Part | What it does | Where to set it |
|---|---|---|
| Persona | Name, description, tone, intro message — how the agent presents itself | General tab |
| Knowledge | Files, projects, URLs, YouTube transcripts the agent can cite | Knowledge tab |
| Tools | What the agent can do — search, browse, send messages, call APIs | Tools tab |
| Model | Which frontier model powers reasoning (Auto picks for you) | General → AI Model |
| Memory | What the agent remembers about each user across sessions | Auto-managed |
| Commands | Reusable /slash shortcuts that trigger pre-written prompts |
Commands tab |
Creating Your First Agent
Building an agent takes about two minutes. There are four ways in — pick whichever matches your starting point.
- Open the Agents tab in any workspace or Genesis app.
- Choose a creation method:
- Generator — describe the agent in plain English (AI Agent Generator)
- Taskade EVE — ask the built-in assistant to create it for you (Taskade EVE guide)
- Template — start from a pre-built agent (Support, Research, Sales, PM, Content)
- Scratch — fill out General → Knowledge → Tools → Commands manually
- Fill out the General tab — name, description, persona, tone, introduction.
- Train it in the Knowledge tab by uploading files, linking projects, adding URLs, or pasting YouTube videos.
- Enable tools in the Tools tab — start with web search and one integration (Slack, Gmail, Notion).
- Add slash commands for the 3-5 tasks you'll repeat most often.
- Click Update and open the Chat tab to test it.
Need prompt patterns? Read the Guide to Writing Agent Prompts.
Choosing the Right Model
Taskade agents can run on 11+ frontier models from OpenAI, Anthropic, and Google. Auto mode is the default and picks the best model for each request — leave it alone unless you have a specific reason to override.
| Mode | Best for |
|---|---|
| Auto (default) | General use. Taskade picks the right model for each turn. |
| Standard | Quick questions, simple drafting, high-volume chat. |
| Thinking | Complex analysis, coding, long-context reasoning. |
| Reasoning | Multi-step problem-solving and creative work. |
| Advanced | Pin a specific model when you need exact control. |
Higher-intelligence models use more credits per interaction. See the AI Usage & Credits guide for details.
Built-in Tools
Every Taskade agent ships with 22+ built-in tools organized into a few clear categories. Tools are how an agent moves from "answering" to "doing."
| Category | Examples |
|---|---|
| Search & retrieval | Workspace search, knowledge lookup, semantic search |
| Web | Web browsing, web search, URL fetching, scraping |
| Project actions | Create projects, add tasks, update status, assign owners |
| File operations | Read files, generate files, summarize uploads |
| Communication | Slack, Gmail, Discord, Teams messaging |
| Code & data | Code execution, calculator, data parsing |
| Integrations | 100+ connected apps via the Integrations Directory |
See the full Tools for AI Agents catalog for the complete list and configuration options.
Custom Tools via MCP
Taskade supports Model Context Protocol (MCP) connectors, so you can plug any MCP-compatible tool or data source into your agent — no custom integration code required.
- Hosted connectors — managed by Taskade, zero setup
- Bring your own server — point at any MCP endpoint
- Stateless mode — lightweight, scalable for high-volume use
Full walkthrough: MCP Connectors guide.
Knowledge & Persistent Memory
Knowledge is what the agent knows. Memory is what it remembers.
- Knowledge is uploaded once and indexed: files, projects, URLs, YouTube transcripts, Google Drive / Dropbox / Box files, Media Manager assets.
- Persistent memory is automatic — the agent remembers each user across sessions, building context over time without you wiring anything up.
The more your workspace grows, the smarter your agents get. Train them once, and every new project becomes additional context for free.
Deep dive: Agent Knowledge & Memory.
Slash Commands and Quick Actions
Slash commands turn your best prompts into one-keystroke shortcuts. Type / anywhere — chat, project, or task — and your custom commands appear.
- Open the Commands tab inside an agent.
- Click ➕ New command.
- Set:
- Name — what users type after
/ - Prompt — the instructions the agent runs
- Mode — Default (one-shot) or Plan & Execute (sets goals, then runs them)
- Name — what users type after
- Save — the command is now available across the workspace.
You can also invoke commands from the Add-Ons menu on any task or paragraph, or run them in bulk by multi-selecting tasks.
Multi-Agent Collaboration
A single agent is great for one job. Multi-agent teams are how you handle workflows that need different specialists — research, then writing, then QA.
Build a team in the AI Teams tab, add your agents, and chat with all of them at once. Each agent contributes its specialty; Taskade routes the conversation.
See the AI Teams guide and Multi-Agent guide.
Embedding Agents Publicly
Any custom agent can be published as a public web app — no auth, just a link.
- Open the agent and choose the Publish tab.
- Customize theme, background, intro message, and meta info.
- Open the Share tab and toggle public access on.
- Copy the link, or grab the embed snippet for your website.
Advanced sharing controls include password protection, chat timeout, knowledge copy on/off, branding removal, and light/dark theming.
Full guide: Share & Embed AI Agents.
Agent + Automation Pattern
Agents and Automations work two directions:
- Automation → Agent: A flow uses Ask Agent or Run Agent Command as a step, letting any trigger (form submit, schedule, webhook) call your agent.
- Agent → Automation: Your agent uses an automation as a tool, calling it inside a conversation when it decides the user needs that action.
The second pattern is powerful: instead of building one rigid flow per task, you wrap each action as a tool and let the agent decide when to use it.
- Build a flow with the Agent Tool trigger.
- Open your agent → Tools tab → enable the automation.
- Choose Manual approval (agent asks first) or Automatic (agent runs it).
See the Automation Guide and Agent Automation.
Pricing & AI Credits
Custom agents are included in every Taskade plan, including Free. AI usage is metered in credits, not seats — every conversation, tool call, and command consumes credits based on the model.
| Plan | Credits | Best for |
|---|---|---|
| Free | 3,000 one-time | Trying agents, personal projects |
| Starter ($6/mo) | Monthly refill | Solo creators and side projects |
| Pro ($16/mo) | Higher monthly refill, 10 users | Small teams, daily use |
| Business ($40/mo) | Largest monthly refill | Growing teams, public agents |
| Enterprise | Custom | Org-wide deployment |
Higher-intelligence models cost more credits per turn. Auto mode optimizes spend automatically. See the AI Usage & Credits guide.
Common Questions
How is a custom agent different from Taskade EVE?
Taskade EVE is the built-in assistant that helps you build other agents and operate Taskade itself. Custom agents are the specialized teammates you create for specific jobs — support, research, sales, content, etc.
Can agents act on my projects automatically?
Yes. Enable project-action tools, and the agent can create projects, add tasks, update status, and assign owners. Pair with automations for fully hands-off workflows.
Do agents remember previous conversations?
Yes. Persistent memory is per-user and survives across sessions. Each user gets their own memory store inside the agent.
Can I use my own AI provider or API key?
Taskade routes through 11+ frontier models from OpenAI, Anthropic, and Google — managed for you. For custom external tools or data, use MCP connectors.
Can I share an agent without giving access to my workspace?
Yes. Publish it from the Share tab — public users get a hosted chat URL with no workspace access. You control whether knowledge is shared.
How do agents fit into Genesis apps?
Every Genesis app can embed one or more custom agents as its intelligence layer — answering user questions, triggering flows, and reading from app data. See Build your first Genesis app.
Related guides
- AI Agent Generator — Text-to-agent in seconds
- Tools for AI Agents — Full 22+ built-in tool catalog
- Agent Knowledge & Memory — Train agents on your data
- Guide to Writing Agent Prompts — Prompt patterns that work
- Multi-Agent Teams — Coordinate specialists on shared goals
- AI Teams — Team-level agent collaboration
- AI Agent Chat — Chat modes and command invocation
- Share & Embed Agents — Publish agents to the web
- Agent Automation — Trigger agents from flows
- MCP Connectors — External tools via Model Context Protocol
- Taskade EVE Assistant — The built-in meta-agent
- AI Usage & Credits — How credits and models price out