Agents
Compose a custom AI agent — personality, expertise, tools, knowledge, and guardrails — then launch it for private use or share it in the Marketplace.
Agents
Build, Launch, and Run Your Own AI Agents
CubDen's Agents system lets you compose a fully custom AI agent — its personality, expertise, integrations, capabilities, knowledge, and guardrails — then launch it for private use or share it in the public Marketplace. Every agent is yours to manage, edit, and deploy across chat, scheduled tasks, and long-running plans.
Personality
Define Who Your Agent Is
Every agent starts with an identity. Give it a name, an intro message, a system prompt, an avatar, and a voice — the persona it brings to every conversation.
Name and intro message
Set how your agent introduces itself in chat. The intro greets users when they open a new conversation.
System prompt
Write the underlying instructions that shape tone, behavior, and decisions across every reply.
Avatar image
Upload a custom avatar or use the default Cub portrait. Crop and preview before saving.
Voice persona
Pick from multiple voice personas so your agent sounds the way you want during voice mode and read-aloud.
Expertise
Focus the Agent's Domain
Select up to three expertise domains from a preset catalog or add your own. Narrow expertise yields higher efficiency — if you need more experts, create more agents.
Preset domains
Start from a curated catalog of expertise domains spanning support, sales, research, engineering, and more.
Custom expertise
Add bespoke domains — quantum computing, regional regulations, internal product lines — that don't exist in the presets.
Expertise notes
Free-form notes for specialized knowledge, preferred frameworks, or niche skills the agent should always apply.
Up to three domains
A focused agent outperforms a generalist. The builder caps expertise at three to keep behavior sharp and predictable.
Integrations
Connect Real-World Tools
Wire your agent into the systems it needs to take real action. Each integration carries scoped permissions and a fallback path if the connection isn't available.
Web Search
Fetch real-time information, news, and documentation directly from the web during a conversation.
Read inbox, send mail, and manage drafts on behalf of the user with explicit permissions.
Calendar
View, create, and manage calendar events and scheduling across connected calendars.
REST APIs
Make HTTP requests to external APIs and process their responses inside an agent run.
Messaging
Send and read messages in Slack, Teams, or other chat platforms the agent has been authorized to use.
File Management
Read, write, and organize files in connected storage like Google Drive or S3 buckets.
Capabilities
Turn On Cognitive Abilities
Capabilities are the foundational powers your agent reasons with — beyond the tools it can call. Toggle each one and customize the behavior description that drives it.
Autonomous Reasoning & Planning
Break complex objectives into subtasks, evaluate trade-offs, and update plans as new information arrives.
Persistent Memory
Retain context across sessions — user preferences, past decisions, project history, and learned patterns.
Tool Use & Integration
Reach for browsers, code interpreters, email, and APIs to perform real-world actions, not just generate text.
Data Analysis & Decision-Making
Process large volumes of data to optimize, detect anomalies, identify trends, and surface actionable insight.
Image Generation
Render images on demand inside chat via /imagine. Results save to the user's AI media library.
Adaptive Learning
Analyze feedback and operational data to refine strategy over time. Coming soon to the capability catalog.
Multi-Agent Collaboration
Coordinate with other specialized agents, delegate subtasks, and synthesize results. Coming soon.
Knowledge
Ground the Agent in Facts
Attach knowledge entries that the agent treats as ground truth. Each entry is classified so the agent knows how to use it during a conversation.
Procedural
Rules, SOPs, policies, and workflows that define how the agent should behave step-by-step.
Declarative
Factual information such as product catalogs, pricing tables, FAQs, and inventory data.
Contextual
Customer records, interaction history, and session data that enable personalized responses.
Unstructured
Raw information lifted from PDFs, manuals, emails, and support tickets the agent should reference.
Guards
Set the Guardrails
Guards are the rules your agent must follow. They constrain output, restrict actions, and ensure the agent escalates instead of guessing when stakes are high.
No Hallucination
Never fabricate facts, data, citations, or URLs. If uncertain, say so explicitly rather than guessing.
Stay In Scope
Only respond to topics inside the agent's defined expertise and role. Politely redirect off-topic requests.
No Harmful Content
Never generate violent, illegal, discriminatory, sexually explicit, or deceptive content.
Data Privacy
Never expose, log, or transmit sensitive PII, credentials, or financial details without explicit authorization.
No Unauthorized Actions
Never execute destructive, irreversible, or high-risk actions without explicit user confirmation.
Human Escalation
Escalate to a human when confidence is low, the request is outside capability, or the user asks for one.
Rate Limiting
Throttle requests and tool calls so the agent stays within safe usage bounds across long-running tasks.
Audit Trail
Record every action the agent takes so its decisions and tool calls can be reviewed after the fact.
Launch
Review, Choose Visibility & Launch
The final step rolls everything up into a single review: personality, expertise, integrations, capabilities, knowledge, and guards. Pick visibility and launch.
Private agents
Only you can view, run, and edit a private agent. Use this for personal workflows and internal experiments.
Public agents
Public agents appear in the Marketplace. Everyone can view personality, expertise, capabilities, knowledge, and guards — but integrations stay private.
Edit any section after launch
Every step is editable from My Agents. Update personality, swap integrations, or tighten guards without rebuilding the agent.
Activate to use anywhere
Launching flips the agent from draft to active so it can be selected in chat, scheduled tasks, and long-running plans.
Marketplace
Discover Public Agents
The Agent Marketplace is the public catalog of agents the CubDen community has built and shared. Browse, preview, and save any public agent to your own account.
Browse and search
Search public agents by name and explore their personality, expertise, capabilities, knowledge, and guards before adopting them.
Preview an agent
Open any agent's detail page to inspect what it does, how it behaves, and what it has permission to use.
Built by the community
Agents are submitted by other CubDen users. Public visibility exposes everything except private integration credentials.
Adopt with one click
Save a public agent into your own My Agents list and start using it immediately in chat or scheduled tasks.
My Agents
Manage Everything You've Built
My Agents is the management console for every agent on your account. View drafts, edit configuration, switch visibility, and assign agents to organizations.
Drafts and live agents
See every agent you've created, with status indicators for draft, active, and inactive states at a glance.
Edit any section
Jump back into the builder at any step — personality, integrations, knowledge — and update without restarting.
Assign to an organization
Move an agent into one of your organizations so teammates inside that org can use it alongside their own.
Switch visibility
Flip an agent between private and public any time. Switching to private removes it from the Marketplace.
Use
Use Agents Anywhere on CubDen
Once an agent is active, it shows up across the platform. Mention it in chat, hand it a scheduled task, or let it drive a long-running plan from end to end.
Mention agents inline
Type @ in any AI chat to pull a specific agent into the thread mid-conversation. Autocomplete finds it as you type.
Pick an agent for a new chat
Start a fresh conversation by selecting from the carousel of agents you've built or saved from the Marketplace.
Run scheduled tasks
Assign any agent to a scheduled task so it executes recurring work — daily digests, weekly reports, monthly invoices — on its own.
Drive long-running plans
Hand an agent a multi-step objective and let it plan, request approval, and execute the run step by step.
Requires a CubDen account
Agents are tied to your account so your configuration, integrations, and knowledge stay private and in sync across devices.