Features and Components
The CAIPE UI is where people use CAIPE day to day. It brings chat, Agent Builder, Skills, Knowledge Bases, credentials, workflows, and administration into one signed-in workspace.
Start with the task you need
- Ask for help: open Chat, choose an agent, and send a request.
- Create a reusable assistant: open Agents, choose Create Agent, and follow the Agent Builder guide.
- Give an agent access to trusted information: create or search a source in Knowledge Bases, then attach it in Agent Builder.
- Coordinate several steps: open Workflows and add agents in order.
- Reuse a procedure: browse Skills, review its scan status, and attach it to an agent.
- Connect an existing web product: open Apps when an administrator has enabled an External App.
Main Surfaces
| Surface | Purpose |
|---|---|
| Chat | Conversation history, streaming responses, feedback, and file/tool timelines |
| Agents | Create and manage Dynamic Agents, models, MCP servers, subagents, and workflows |
| Skills | Browse, import, run, scan, and revise reusable skill templates |
| Knowledge Bases | Ingest sources, run search, and inspect graph data when RAG is enabled |
| Credentials | Connect user and service credentials for MCP-backed tools |
| Admin | Platform settings, users, teams, RBAC, audit, metrics, and health |
Global Search and Command Palette
- Select the Search control in the top-right application header, or press Command+K on macOS and Ctrl+K on Windows or Linux.
- Navigation results are built from the same feature flags and admin tab gates as the sidebar. Hidden or disabled destinations are not added to the index.
- Resource results use existing user-scoped BFF endpoints. Their authorization checks run before data reaches the palette.
- Use Up/Down Arrow to select a result and Enter to open it.
Initial resource coverage:
| Resource | Authorization source | Destination |
|---|---|---|
| Agents | OpenFGA agent#use filtering | Agent editor |
| Conversations | Ownership, sharing, and OpenFGA discovery filtering | Chat thread |
| Skills | Skill catalog OpenFGA read filtering | Skill workspace |
| Workflows | Workflow visibility and OpenFGA read filtering | Workflows |
Deferred resource types:
- Knowledge Base collections: add after the collection page supports a stable item deep link.
- Schedules, MCP servers, and models: add after their list APIs expose user-scoped search contracts and stable item deep links.
- A dedicated search service is intentionally deferred. The palette performs small parallel queries against authoritative APIs and tolerates a partial endpoint failure.
Chat
- Conversations are stored in MongoDB when
MONGODB_URIis configured. - Each conversation targets a Dynamic Agent through
agent_id. - The BFF streams AG-UI/SSE events from Dynamic Agents through
/api/v1/chat/stream/*. - Runtime events render as content, tool calls, warnings, subagents, and status updates in the chat timeline.
Dynamic Agents
- Agents define prompts, model choice, MCP server access, built-in tools, subagents, visibility, and team sharing.
- Admin users can seed config-driven agents and models through Helm app config.
- Agent runs persist checkpoint and file state in MongoDB through the Dynamic Agents backend.
MCP Servers
- MCP server rows can use direct endpoints or AgentGateway-routed endpoints.
- The UI can discover AgentGateway targets and normalize endpoint paths.
- Probe actions verify connectivity and credentials before an agent uses a tool.
Skills
- Built-in templates are read-only by default.
- Users can clone templates into editable copies.
- Skill scan history and safety results are stored in MongoDB.
- Runnable skills launch chat conversations against a resolved Dynamic Agent.
Admin and Health
- Health probes cover the UI/BFF, Dynamic Agents, MongoDB, Keycloak, OpenFGA, AgentGateway, RAG, web ingestor, and migrations when configured.
- Audit surfaces read from the audit service and MongoDB-backed RBAC records.
- Platform settings include default agent/model choices and runtime feature controls.