Agent templates for the fips-agents CLI. Scaffolds production-ready AI agents that deploy to Red Hat AI, communicate with LLMs via the OpenAI async SDK (any OpenAI-compatible endpoint), and let developers focus on prompts, tools, model selection, and evals.
Both templates (agent-loop and workflow) are implemented, along with the shared fipsagents package (on PyPI) and an optional code execution sandbox sidecar.
A developer runs fips-agents create agent my-agent, selects a template variant (agent loop or agentic workflow), and gets a project that compiles, runs locally, and deploys to Red Hat AI via Helm. The scaffolded project includes AI-assisted slash commands (/plan-agent, /create-agent, /exercise-agent, /deploy-agent) that guide development from design through deployment.
The core abstraction is BaseAgent -- a pure Python async class that handles LLM communication (via the OpenAI async SDK), tool dispatch across two planes (agent-code and LLM-callable), MCP client connections (FastMCP v3), prompt loading, skill management (agentskills.io spec), and lifecycle. A typical agent subclass is 20-30 lines.
- docs/ -- Architecture, design decisions, problem statement, and vision.
- planning/ -- Requirements, scope, constraints, and next steps.
- fips-agents/code-sandbox -- Code execution sandbox sidecar (extracted to standalone repo).
Agents built from this template run on Red Hat AI and consume services deployed by rh-ai-quickstart/ai-architecture-charts (vLLM, LlamaStack, PGVector, etc.). The Helm chart in each scaffolded agent bundles only the agent itself.
- fips-agents/gateway-template -- OpenAI-compatible Go gateway that fronts agent deployments
- fips-agents/ui-template -- Minimal chat UI that talks to the gateway
- fips-agents/fipsagents-platform -- Cross-agent platform service for centralized feedback / sessions / traces (multi-agent topologies)
- fips-agents/code-sandbox -- Code execution sandbox sidecar
- fips-agents/mcp-server-template -- Sister template for MCP servers
- redhat-ai-americas/memory-hub -- Optional enterprise memory layer
- agentskills.io -- Skills specification
- agents.md -- AGENTS.md open standard