Specialist agents
Out of the box the hub can run in-process specialists (backend, frontend, platform, security, architecture, and code review personas) driven by ~/.neural-junkie/config.json. Each binds to its own provider and model so you can keep heavy coding on Ollama while routing review work to Claude, for example.
Assistant
Assistant is the always-on helper: product guidance, slash-command help, reminders, tasks, notes, Google Meet note sync, workspace-grounded follow-ups with persistent storage, and a timer-based safety net when human questions go unanswered — the “ops brain” next to your specialists.
Repository expert agents
/create-repo-agent indexes a path you choose, watches for file changes when enabled, and answers with project-local context. Reindex, watch toggles, and lifecycle commands live in the same slash-command surface as everything else.
Confluence documentation agents
/create-confluence-agent attaches a space key so answers cite your wiki. Credentials live in environment configuration; see the Confluence agents guide for indexing and search semantics.
Experts and CLI agents
/create-expert adds specialists on demand: curated presets (backend, frontend, devops, security, architecture, code-review, assistant) or any custom domain — guitar teacher, legal advisor, whatever fits your work. Use a slash command or open a New DM from the desktop with a custom persona. Optional Cursor and Gemini CLI integrations treat external CLIs as first-class providers when binaries and keys are available.
Optional standalone processes
make agents still launches the classic multi-process layout for people who prefer separate OS processes. If you run both in-process and standalone agents with the same names, you can get duplicate registrations — pick one model per environment.