Pick an agent.
Give it a real problem.
Each agent is a configuration of the same engine. Open one and the workspace shows you both halves at once: the conversation, and the execution behind it.
- Organization
- wwwdot
- Plan
- demo
- Model
- claude-opus-5
- Persistence
- Supabase connected
- Agents live
- 4 of 10
- Session ceiling
- 25 messages
Customer Support
Answers from your documentation only, resolves what it can, and escalates with full context when it cannot.
- Grounding
- Docs only
- Escalation
- Context-preserving
- Systems
- Order lookup
Research Analyst
Plans a research approach, gathers sources, extracts what matters and produces a structured report.
- Output
- Cited report
- Method
- Plan-first
- Sources
- Tracked
Internal Operations
A private assistant over company data — reporting, policy lookup and internal tooling for your team.
- Access
- Role-scoped
- Data
- Internal
- Audit
- Full trail
Workflow Automation
Triggered, multi-step execution with conditional logic, API calls and approval gates.
- Trigger
- Event / cron
- Logic
- Conditional
- Safety
- Approval gates
Custom AI SaaS
A complete product: customer accounts, custom workflows, usage metering and billing architecture.
- Tenancy
- Multi-tenant
- Metering
- Usage-based
- Ownership
- Your codebase
Sales Engineer
Qualifies a business problem, grounds every claim in retrieval, and designs a concrete system on the call.
- Tools
- 7
- Knowledge
- 12 docs
- Captures
- Leads
Agent Factory
Tell it what you want built. It works out the agent type, designs it with you, and produces a specification you approve before anything is created.
- Blueprints
- 6
- Output
- Specification
- Deploy
- Studio
Order Help Desk
A customer support agent that answers questions about order status, shipping times and returns using your help centre articles, refuses anything about payment disputes, and escalates to a human support agent when it cannot answer from the articles.
Front Desk Assistant
Answers hotel guest questions — check-in/out, amenities, parking, Wi-Fi, breakfast, restaurant hours, room, cancellation, pet and accessibility policies, directions — strictly from the hotel knowledge base. It never guesses, never books, modifies reservations, processes payments or accesses guest records, and escalates anything else to the front desk.
BTC RSI Trader
A paper-mode trading analysis agent that generates BTC buy/sell signals from RSI thresholds, capped at 50% position size per trade and halting signals at 15% drawdown. All trading capabilities are design-only; no orders are placed.
Tools are declared once for the whole platform and referenced by name from an organization's agent configuration. An agent can only call a tool it declares, and every tool writes through its own tenant — so one client's agent has no path to another client's data.
| Tool | Stage | Purpose |
|---|---|---|
| search_knowledge | knowledge | Search this organisation's knowledge base for factual information about its services, capabilities, approach, policies and process. |
| classify_intent | understand | Classify what kind of agent the user is asking for. |
| plan_agent | plan | Record the plan for designing this agent — the things you need to establish before a specification is possible. |
| search_capabilities | knowledge | Look up what agent blueprints and capabilities this platform offers, including which capabilities are implemented and which are design-only. |
| select_blueprint | knowledge | Lock in the blueprint this agent will be built from. |
| record_configuration | tool | Store the user's answers as structured configuration. |
| draft_agent_specification | decision | Produce the finished agent specification once every required configuration field is answered. |
| build_agent | action | Create the live agent from the approved specification. |
| attach_knowledge_base | action | Attach a knowledge base to an agent that has already been built, so its answers can be grounded in this tenant's documents. |
| test_agent | decision | Verify a built agent against the real runtime before it is published: the model call, its instructions, its tool allow-list, and — when the blueprint requires grounding — that retrieval actually returns content. |
| deploy_factory_agent | action | Publish an agent that has already been built so it becomes usable in Studio. |
| capture_lead | action | Create the lead record for this conversation with whatever details the visitor has volunteered. |
| update_lead | action | Add newly learned details to the existing lead record. |
| calculate_lead_score | decision | Compute a qualification score from the details recorded so far. |
| recommend_solution | decision | Record a concrete solution recommendation once you understand the visitor's workflow well enough to name specific components. |
| create_demo_request | action | Log a request for a tailored demo of the system you recommended. |
| schedule_consultation | action | Register interest in a scoped discovery call. |