AI Technology RadarAI Technology Radar
Trial

A Context Hub is a central, versioned source that supplies agents with the knowledge and conventions they need at runtime - up-to-date API documentation, domain knowledge, project conventions - instead of relying on the model's training cutoff or scattering context across ad-hoc prompts. It is the "provisioning" side of Context Engineering: one curated place agents pull from, rather than each team re-solving context per project.

Building blocks seen in practice:

  • Documentation registries for coding agents that serve current, versioned API docs (e.g. Andrew Ng's Context Hub / chub, Context7) - they close the "model doesn't know the latest library version" gap.
  • MCP resources / servers as the delivery mechanism to any agent.
  • Convention files (AGENTS.md/CLAUDE.md) and neutral knowledge formats such as OKF as the storage layer.

Treat the hub as versioned, reviewed content - it is shared context that also becomes an attack surface if it can be poisoned.