Dashboard
The dashboard is a Next.js app (apps/dashboard-web) served by the cortex-api container on port 4000. It shows what the hub knows and is where people, not agents, make decisions: API keys, LLM providers, user gates.
WARNING
The dashboard has no login. Anyone who can open it can create keys and approve gates. Keep it on a private network or behind an access proxy, and set API_AUTH=required. See Self-host.
Pages
| Page | Path | What is on it |
|---|---|---|
| Dashboard | / | System overview: savings from cortex tools, intelligence, quick connect, recent activity, system resources |
| Organizations | /orgs | Organizations and their projects. Add a repo (GitHub, GitLab, Bitbucket, Azure or local), with an optional token for private repos. Cross-repo search stays inside one organization |
| Project | /projects?id=… | One project: indexing progress, branches, re-index, re-run memory embedding, scan the repo's docs into knowledge, move it to another organization, delete it |
| Knowledge | /knowledge | The knowledge base by project: documents, chunks, search hits, and the files each note is shown for |
| API Keys | /keys | Create, edit and revoke the keys agents and workers use. A new key is shown once |
| LLM Providers | /providers | Providers (API key or OAuth) and the routing of the chat and embedding models, with fallbacks |
| Usage | /usage | Token use and requests over time, and the savings of cortex tools over 7 days |
| Quality | /quality | Quality reports, scored on four dimensions |
| Conductor | /conductor | Plans as trees, live agent cards, task details, Needs you, user gates, Delete pipeline |
| Sessions | /sessions | Agent sessions and handoffs, with their details |
| Settings | /settings | Endpoints and remote access, MCP config to copy, how to run a worker, LLM switches, hub configuration, notification preferences, Docker services, environment |
| Setup | /setup | The first-run wizard: welcome, provider sign-in, model selection |
What agents see of it
Everything on the dashboard comes from the same API the MCP tools use. An agent's cortex_quality_report shows on Quality, its session on Sessions, its knowledge on Knowledge, its Conductor tasks on Conductor. Only people approve user gates; MCP calls are refused.