template-agent merges Deep Agent into main

August 11, 2026

The Agent template now builds on LangGraph Deep Agents with orchestrator, subagents, skills, and enterprise infrastructure.

The template-agent repository has merged its deep-agent development branch into main (PR #104). This is the largest architectural change since the template launched.

What changed

The template is no longer a simple LangGraph script with embedded prompts. It is now a Deep Agent runtime built on LangGraph Deep Agents and the Aegra CLI.

Agent capabilities:

• Orchestrator with analyst and publisher subagents (example domain: fitness assistant)
• Skills: client-intake, bmi-report, email-formatter
• MCP integration with SSO pass-through, OAuth, and Dynamic Client Registration (DCR)

Infrastructure:

• Aegra dev server with Redis-backed SSE streaming on port 5002
• PostgreSQL for checkpoints, memory, and feedback storage
• Langfuse tracing and OpenTelemetry metrics
• OpenShift and Kind deployment overlays

What stayed the same

The three-repo model is unchanged: template-agent, template-mcp-server, and template-ui still work together as a stack. MCP and UI remain separate repositories.

What this means for you

• Clone fresh if you forked the old template — the deep_agent/ package and config/agent/ layout replace the previous template_agent/ structure.
• Use make local instead of python -m agent_template — the agent runs via Aegra on :5002.
• Set Vertex AI credentials in .env (GOOGLE_APPLICATION_CREDENTIALS_CONTENT) before first run.
• Read the config-as-code announcement for how prompts and subagents are defined without Python edits.

Next: Config-as-code architecture · Agent Template docs (updating)