How do I use Ollama for code completion in VS Code?

Install the Continue.dev extension, then add a model with provider: ollama and apiBase: http://localhost:11434 to ~/.continue/config.yaml . Pull a coder model first and restart the editor.

Which local model is best for autocomplete?

A small, fast coder model like qwen2.5-coder:1.5b gives near-instant tab suggestions. Use a larger model such as qwen2.5-coder:7b for chat, explanations, and refactors where quality matters more.

Can I use Ollama with Cursor?

Partly. Cursor needs an OpenAI-compatible URL, but it routes requests through its own cloud, so a local server must be exposed with a public tunnel. Continue.dev is the better fully-local option.

Does connecting an IDE to Ollama use the Language Server Protocol?

No. Extensions like Continue.dev call Ollama's HTTP API directly at http://localhost:11434 . There is no LSP involved, so you only configure a base URL and a model name.

Build / AI / Ollama

Ollama In Your IDE

Connect editor tools like Continue.dev to a local Ollama server for private, free code completion.

TL;DR

Point IDE extensions at the local 11434 port.
Pick a coder model like qwen2.5-coder for autocomplete.
Set apiBase to localhost:11434 in the extension config.

Jump to section

Prepare Ollama
Continue.dev Config
Cursor And Others
Pick A Model
Tips
Warnings
In Practice
FAQ

Prepare Ollama

Pull A Coder

Download a coding model for the editor.

ollama pull qwen2.5-coder

Keep It Loaded

Pin the model so autocomplete stays fast.

export OLLAMA_KEEP_ALIVE=-1

Confirm It Runs

Check the server responds on its port.

curl http://localhost:11434

Continue.dev Config

Model Entry

Add a model block to ~/.continue/config.yaml.

models:
  - name: Ollama Coder
    provider: ollama
    model: qwen2.5-coder
    apiBase: http://localhost:11434

provider

Tell Continue to use the Ollama provider.

provider: ollama

apiBase

Point the block at the local server URL.

apiBase: http://localhost:11434

Cursor And Others

OpenAI Base URL

Cursor needs an OpenAI-compatible endpoint.

# Set base URL to the /v1 path in settings

Public Tunnel

Expose localhost, since Cursor calls from its cloud.

# A tunnel maps a public URL to :11434

Prefer Local Tools

Continue.dev runs fully local, no tunnel.

# No cloud round-trip with Continue.dev

Pick A Model

Autocomplete

A small coder model gives instant suggestions.

ollama pull qwen2.5-coder:1.5b

Chat And Edits

A larger model handles refactors and Q&A.

ollama pull qwen2.5-coder:7b

General Coder

codellama is a solid all-round choice.

ollama pull codellama

Tips

Use a fast small coder model like qwen2.5-coder:1.5b for tab autocomplete, and a larger one for chat and refactors.
Keep ollama serve running and pin the model with OLLAMA_KEEP_ALIVE=-1, so suggestions stay instant while you type.

Warnings

Continue.dev talks to Ollama over its HTTP API, not the Language Server Protocol; there is no LSP bridge to configure.
Cursor proxies requests through its cloud, so a localhost Ollama needs a public tunnel or it cannot be reached.

In Practice

Pull a coder model, keep it loaded, and point Continue.dev at the local server in config.yaml.

A small coder model keeps tab suggestions near-instant.
Pinning it in memory avoids a reload pause mid-typing.
The config block routes Continue to the local server.
A restart loads the config; no LSP or cloud key is needed.

# 1. Pull a fast coder model
ollama pull qwen2.5-coder:1.5b

# 2. Keep it loaded for instant suggestions
export OLLAMA_KEEP_ALIVE=-1

# 3. Add this to ~/.continue/config.yaml:
#    models:
#      - name: Ollama Coder
#        provider: ollama
#        model: qwen2.5-coder:1.5b
#        apiBase: http://localhost:11434

# 4. Restart the editor, then verify
curl http://localhost:11434