Custom Agents

Build your own team-owned agents with custom system prompts, model selection, tool access, skills, subagents, and workflows. Use the no-code Agent Builder UI for day-to-day authoring, or bootstrap config-driven agents with the CAIPE UI chart.

Quick links: Helm Chart Docs · Developer Guide

---

Dynamic Agents Service

Dynamic Agents is the chat runtime and agent-builder service.

• Each custom agent gets its own system prompt, model, tool access, and persona
• MCP tool support: connect registered MCP servers over stdio or streamable http
• Built-in tools include URL fetch, current datetime, user info, wait, and approved workflow execution
• REST + SSE API for browser, bot, workflow, and service callers
• Deploy via Helm: oci://ghcr.io/cnoe-io/charts/dynamic-agents

Agent Builder UI

No-code agent creation directly from the CAIPE web UI.

• Configure owner team, visibility, system prompt, model, and allowed tools interactively
• Attach registered MCP server tools and built-in tools to each agent
• Add skills, subagents, middleware settings, human approval rules, and workflow access
• Instantly test agents in the chat UI without redeployment
• Use team or global visibility with RBAC-controlled access

App Config — Pre-wire Models, MCP Servers, Agents, and Workflows

Bootstrap CAIPE resources at chart install time using appConfig in the caipe-ui chart. Bootstrapped resources are marked config_driven and cannot be edited through the UI.

Key	Purpose
appConfig.models	Available LLM endpoints for agent selection
appConfig.mcp_servers	MCP server endpoints to register at startup
appConfig.agents	Agent definitions: name, system prompt, model, tools, skills, subagents, and visibility
appConfig.workflow_configs	Workflow definitions that agents can trigger when granted access

Idempotent: safe to apply on upgrades without duplicating entries. Removed config-driven entries are cleaned up on the next startup.

Customizable System Prompts

• Per-agent system prompts for Platform Engineer, SRE, Developer, or service-specific personas
• Ready-to-use SRE agent included out of the box — customize prompts, tools, and workflows for your team
• Prompt changes are saved in MongoDB for UI-managed agents; config-driven agents keep prompts in Helm values for GitOps review
• Workflow-enabled agents receive an automatic prompt addendum listing approved workflows

Multi-Model Support per Agent

Each custom agent can target a different configured model/provider entry.

• Define available models in appConfig.models
• Select or change an agent's model in the Agent Builder UI
• Use OpenAI-compatible provider entries, including LiteLLM-backed endpoints
• LLM secrets managed via Kubernetes Secrets or ExternalSecrets Operator

Production Deployment

• Kubernetes Helm chart with configurable replicas, resources, and HPA
• AGENT_RUNTIME_TTL_SECONDS — auto-expire idle agent runtimes
• MongoDB-backed persistence for agent state and conversation history
• ExternalSecrets integration for secrets management in GitOps workflows
• Prometheus metrics endpoint at /metrics for observability