CAIPE Architecture Evolution
This document tracks the architecture evolution roadmap for the Cisco AI Platform Engineering (CAIPE) system, from the current static distributed model through dynamic agent unification and persona-based profiles. Pulled from PR #975 for spec 093 (093-agent-enterprise-identity).
Roadmap
Phases
Phase 1: Current — Static Distributed Agents
The baseline CAIPE architecture.
- Remote distributed agents and MCP servers
- Customized via Agent Registry and Helm config
Phase 2: March 2026 — Single Node and Dynamic Agents
Consolidation and extensibility in parallel.
| Feature | Description |
|---|---|
| Static Single Agent | Single node architecture for more efficient agent communication and tracing; continue remote BYO agent support |
| Task Config | Support for deterministic task configuration |
| Skills Integration | Skills middleware with CAIPE supervisor |
| Separate Dynamic Agent | Users can create their own dynamic agents with custom personas and chosen MCP tools |
Phase 3: March 2026 — Security and Governance
Cross-cutting concerns that apply to all agent types.
| Feature | Description |
|---|---|
| RAG Authorization | Per knowledge-base RBAC; user context authorization |
| I/O Guardrails | Slackbot input/output compliance guardrails |
| Per Agent LLM Budget | Per agent LLM configuration and budget controls |
Phase 4: April 2026 — Unification
| Feature | Description |
|---|---|
| Dynamic/Single Unification with BYO Remote | Unify the static single agent and dynamic agent architecture |
Phase 5: May 2026 — Persona-based Profiles
| Feature | Description |
|---|---|
| Persona-based Dynamic Agent Profiles | Support team-based default dynamic agents; team/user-based MCP tool access |