Stack recipe · Reviewed June 2026

Local Coding Assistant Stack

A coding stack for testing open coding assistants with local or self-hosted models, repository-aware tools, and human code review.

Bill of materials

IDE extension

Continue →

VS Code + JetBrains extension — inline edit, chat, and repo-context indexing with local models

Cline →

Autonomous coding agent for VS Code; reads, writes, and runs terminal commands with approval

Model runner

Ollama →

Local inference API — set OLLAMA_HOST=0.0.0.0:11434 to expose on your LAN for IDE access

LM Studio →

GUI runner with OpenAI-compatible server; useful for switching models without a terminal

Recommended coding models

Qwen3 8B →

Strong code generation and tool-call reasoning; fits on 8 GB VRAM at Q4_K_M

DeepSeek R1 14B →

Chain-of-thought reasoning; excellent at explaining diffs and debugging logic

Mistral Small 3.1 →

Reliable general fallback when coding specialists miss broad context

Set up Ollama + Continue in minutes

1. Install Ollama and pull a coding model

curl -fsSL https://ollama.ai/install.sh | sh && ollama pull qwen3:8b

2. ~/.continue/config.json (macOS/Linux) or %USERPROFILE%\.continue\config.json (Windows)

Connect Continue to your local Ollama instance. Reload VS Code after saving.

{
  "models": [
    {
      "provider": "ollama",
      "model": "qwen3:8b",
      "title": "Qwen3 8B (local)",
      "apiBase": "http://localhost:11434"
    }
  ],
  "tabAutocompleteModel": {
    "provider": "ollama",
    "model": "qwen3:8b",
    "apiBase": "http://localhost:11434"
  },
  "contextLength": 32768
}

3. Verify Ollama API is reachable

curl http://localhost:11434/api/tags

4. Install Continue extension (VS Code)

code --install-extension Continue.continue

Ran the commands? Measure what your setup actually does → — speed, load time, and whether anything is silently throttling it.

Best for

Developers who want coding help close to their editor while keeping model choice, context, and review workflows under control.

Core tools

Continue
Aider
Ollama
LM Studio
OpenCode

Recommended models

Qwen3 Coder
DeepSeek Coder V2
DeepSeek R1 distills
Small Qwen or Mistral variants for local tests

Hardware notes

Small coding models can run on many developer machines. Larger coding models need more VRAM or a hosted/self-hosted inference endpoint.

Setup steps

Pick one assistant interface for your normal coding workflow.
Connect it to a local runtime or a controlled model endpoint.
Test on a small repository task with clear expected changes.
Review every diff before applying it.
Track which model works best for your language, framework, and repo size.

Trade-offs

Local coding assistants can reduce data exposure but may be slower or weaker than frontier hosted models. Generated patches still need careful review.

Alternatives

Use Continue for IDE-native workflows.
Use Aider for terminal-first git-aware edits.
Use a hosted model when quality matters more than local control.

Related resources

Continue tool profile Aider tool profile Ollama tool profile Open-source AI coding assistants Continue vs Aider comparison

Not sure if your PC has enough VRAM for this workflow?

Run the Local LLM Hardware Checker →

FAQ

Can local coding assistants edit files automatically?

Some can, but you should keep review steps in place and avoid unattended writes until the workflow is proven.

Which model should I test first?

Start with a coding-focused model that fits your hardware, then compare it on real tasks from your own repositories.

Get practical stack updates

Join the OpenSourcesAI update list for new stack recipes, tool notes, and developer-first comparisons.

Get updates Browse guides