Skip to main content
Version: main 🚧

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​

SurfacePurpose
ChatConversation history, streaming responses, feedback, and file/tool timelines
AgentsCreate and manage Dynamic Agents, models, MCP servers, subagents, and workflows
SkillsBrowse, import, run, scan, and revise reusable skill templates
Knowledge BasesIngest sources, run search, and inspect graph data when RAG is enabled
CredentialsConnect user and service credentials for MCP-backed tools
AdminPlatform 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:

ResourceAuthorization sourceDestination
AgentsOpenFGA agent#use filteringAgent editor
ConversationsOwnership, sharing, and OpenFGA discovery filteringChat thread
SkillsSkill catalog OpenFGA read filteringSkill workspace
WorkflowsWorkflow visibility and OpenFGA read filteringWorkflows

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_URI is 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.