ai_hub is a reusable Django app for configuring, operating, and auditing AI agent systems from Django Admin.

It gives a host project an admin-driven AI operations layer where staff can define providers, models, agents, tools, knowledge, fixed workflows, autonomous sessions, and execution telemetry without scattering AI logic across views, tasks, and scripts.

AI features become hard to maintain when prompts, provider calls, tool access, execution rules, and debug logs are hidden inside product code.

ai_hub separates those concerns:

• product code owns the business domain,
• ai_hub owns reusable AI configuration and execution primitives,
• admin users can inspect how AI is configured,
• developers can integrate AI workflows without rebuilding the platform layer each time,
• failed runs leave inspectable traces instead of disappearing into logs.

ai_hub has two workspaces.

The workspaces are visually separated in Admin, but they share the same technical base:

• providers,
• models,
• agents,
• knowledge,
• tools,
• execution telemetry.

ai_hub supports:

• provider configuration,
• model configuration,
• agent profiles with prompts and contracts,
• knowledge collections and documents,
• tool definitions,
• sequential pipeline orchestration,
• autonomous GAME goals (durable goals, deterministic scheduling, action dispatch, scoped memory, approval/resume, policies, plans, and sub-agent delegation),
• per-capability GAME feature flags,
• staff-only execution endpoints,
• execution sessions and step logs,
• admin control center metrics and operational GAME dashboards,
• guided forms and examples.

ai_hub should not own:

• public product UI,
• billing,
• product-specific database objects,
• final domain persistence,
• user-facing business flows,
• private secrets.

Those responsibilities belong to the host project.

The Admin experience is part of the product.

ai_hub should feel like an operational backend, not only a raw Django model list. It uses:

• an AI Hub home dashboard,
• separate Orchestrator and GAME workspaces,
• shared resource sections,
• a visual connection graph,
• recent sessions and health metrics,
• guided changelists,
• guided forms with examples and placeholders,
• readable execution timelines.

The design goal is that a non-expert admin can understand what to create next and why.

If a feature only makes sense for one product domain, keep it in the host project or behind a small adapter.

If a feature helps any AI-enabled Django project, it belongs in ai_hub.