Skip to content
← Tools / Skills
ToolContinue Dev/10 trustMetrics-checked

Continue Hub

A registry for creating and sharing custom AI coding assistants — Docker Hub for agents.

Visit official site →

About

Launched alongside Continue 1.0 in February 2025 (backed by a $3M seed round), Continue Hub is the sharing layer for the open-source Continue AI coding assistant: a registry where developers and vendors publish custom AI coding assistants and the composable building blocks they're made from — models, context providers, prompts, and rules. Continue pitches it as Docker Hub, GitHub, or Hugging Face for coding assistants.

At launch the hub shipped with pre-built assistants plus verified-partner blocks including Mistral's Codestral, Anthropic's Claude 3.5 Sonnet, and DeepSeek-R1 via Ollama, with any vendor or individual able to contribute. It complements Continue's core value proposition — the leading open-source (Apache 2.0), model-agnostic AI code assistant for VS Code and JetBrains, runnable against air-gapped local inference. Honest caveats: community-contributed assistants vary in quality and maintenance, enterprise teams should vet blocks before org-wide rollout, and the hub's long-term catalog depth depends on sustained community contribution.

What it can do

  • Registry of pre-built custom AI coding assistants
  • Shareable blocks: models, context providers, prompts, rules
  • Verified partner blocks (Mistral, Anthropic, DeepSeek/Ollama)
  • One-click assistant install into VS Code and JetBrains
  • Team sharing of standardized assistant configs

Setup

Install the Continue extension for VS Code or JetBrains, then browse and pull assistants from the hub.

Prerequisites

Continue extension installed; API keys or local model endpoints for the assistants you use.

Use cases

ai-agentscode-assistantmarketplaceopen-source

Works with

Trust score — 7/10

/10 trustMetrics-checked

Scored 7/10 (metrics tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 70; Trending score 70/100 from researcher assessment.

Sources

Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.