Continue.dev

Continue is an open-source AI assistant for VS Code, JetBrains, and Neovim. It ships chat, inline edits, autocomplete, codebase-wide semantic search, and an agent mode that plans and runs multi-step tasks (file edits, tool calls) inside the IDE. Continue also has a separate product line for source-controlled AI checks on pull requests via the cn CLI — that side runs in CI, not the IDE, and isn't covered here. Model assignment in the IDE extension is role-based: each model entry in your YAML config picks up one or more of chat, edit, apply, autocomplete, embed, rerank, summarize, so autocomplete can run on a fast small model while chat and agent runs on something stronger. Configured through a single YAML file with first-class support for OpenAI-compatible providers, which is all Melious needs. We recommend splitting Continue across at least two Melious models — one small for autocomplete, one strong for chat / edits / agent. One model doing both ends usually disappoints somewhere: too slow at the keyboard, or too dumb in the sidebar.

Setup

name: Local Assistant
version: 1.0.0
schema: v1

models:
  - &melious
    name: Melious · Chat
    provider: openai
    apiKey: sk-mel-<YOUR_API_KEY>
    apiBase: https://api.melious.ai/v1
    model: glm-5.1
    roles: [chat, edit, apply]

  - <<: *melious
    name: Melious · Autocomplete
    model: glm-5.1
    roles: [autocomplete]

Role-based routing

Continue lets you assign different models to different tasks via the roles field:

Autocomplete wants latency; chat and edit want capability. Split them across two model entries — a small/fast model on [autocomplete] and a strong one on [chat, edit, apply]. glm-5.1 works for both as a starting point.

Embeddings

If you use Continue's indexed codebase search, it runs embeddings through whichever provider is configured for the embed role. Our /v1/embeddings is OpenAI-compatible, so the same provider: openai config works — just add another model entry reusing the anchor:

models:
  - <<: *melious
    name: Melious · BGE M3
    model: bge-m3
    roles: [embed]

bge-m3 is a good default for mixed-language codebases. The API key you use needs the inference.embeddings scope. See Models for alternatives.

Reranking — not yet supported through Continue

Melious's /v1/rerank is Cohere-compatible, not OpenAI-compatible. Continue routes the rerank role through one of: cohere, voyage, huggingface-tei, or LLM-rerank via openai (slow + inaccurate). Continue's cohere provider hard-codes an allow-list of Cohere model names, so it won't accept bge-reranker-v2-m3 against a custom apiBase. Until we either (a) ship a TEI-shaped rerank endpoint or (b) Continue's cohere provider drops its model allow-list, codebase rerank with Melious models doesn't work through Continue.

Workaround if you really need it: skip the rerank role — embeddings alone produce decent retrieval, and the latency win from skipping rerank is non-trivial. Or fall back to LLM-rerank with provider: openai pointed at a small Melious chat model:

models:
  - <<: *melious
    name: Melious · LLM Reranker (fallback)
    model: glm-5.1
    roles: [rerank]

This is slower and less accurate than a real reranker. Treat it as an interim option, not a recommendation.

What's different

• EU-residency on every role. Chat, edit, autocomplete, and embed calls all stay in Melious's EU provider network. Continue's local indexer stays on your machine. Rerank isn't wired up yet — see the section above.
• Image inputs work for models that support vision. Continue lets you paste screenshots into chat — the request routes image parts through unchanged.
• Autocomplete character count matters. Continue asks for short completions, often fewer than 20 tokens. Small models feel much snappier than large ones for this role — browse melious.ai/hub/models for a fast pick.

When it breaks

• No completions appearing — Continue's statusbar shows which model is the autocomplete target. If it says "not configured", you missed the autocomplete role on a model.
• 401 Unauthorized — apiKey is wrong or rotated. Rotate in our dashboard, paste the new key.
• Laggy chat — the chat model is too heavy. Pick a smaller one from melious.ai/hub/models.

Errors and retry patterns: Errors.