Continue - The AI Coding Tool That Actually Lets You Choose Your Model

Continue runs in VS Code and JetBrains IDEs. It's got 28k stars on GitHub because it solves the one thing every developer hates about Copilot - you're not locked into Microsoft's models. Open source, Apache 2.0 licensed, and you can switch between any AI model you want.

Why Open Source Actually Matters Here

Unlike Copilot (owned by Microsoft) or Cursor (venture-funded startup), Continue is Apache 2.0 licensed. This means you can audit the code, modify it, and you're not fucked if the company changes direction or gets acquired.

Here's why that matters: I use GPT-4 for heavy refactoring, Claude when I need it to actually think, and local Ollama models when I'm working on shit I can't send to the cloud. It also works with Gemini, Azure OpenAI, and Bedrock. Copilot? You get GPT and you'll fucking like it.

The Four Ways Continue Works (And When Each One Breaks)

Continue has 4 modes, each with different success rates:

Agent Mode - Tell it "implement user authentication" and it tries to do the whole thing. Works maybe 70% of the time for simple stuff, 30% if you ask it to do anything complex. When it nails it, you feel like a wizard. When it fucks up, you'll spend longer cleaning up its mess than just doing it yourself. Demo here if you want to see it work properly. Usually it gets confused about context, starts hallucinating imports that don't exist, or gets stuck in a loop trying to fix something that wasn't broken.

Chat - Ask it about your code. "Explain this function" or "why is this query slow" and it actually knows what you're talking about. The context awareness doesn't suck - it reads your actual project files instead of making you copy-paste everything like ChatGPT.

Inline Edit - Highlight code, tell it what to change, boom. This is the feature that actually works. Gets it right about 85% of the time for simple edits. Way better than the copy-paste dance with ChatGPT.

Autocomplete - Tab completion like Copilot, but you pick the model. Quality totally depends on what model you're running. Local models are slow as fuck but keep your code private. Cloud models are fast but send your shit to OpenAI.

Team Setup (And Why Continue Hub Actually Works)

Continue Hub solves the "every developer configures things differently" problem. You can set team-wide model defaults, shared prompts, and API keys without micromanaging individual setups. Unlike most enterprise dashboards, it's actually useful.

Here's the killer feature nobody talks about: MCP tools. Continue can create Linear tickets from your code, read GitLab repos, or grab docs from Confluence without you switching tabs. It's this protocol thing Anthropic built that lets AI tools talk to other services. Continue works with tons of them - databases, files, APIs, whatever. No other coding AI does this shit.

Reality check: Enterprise adoption works because Continue doesn't lock you into one vendor. You can run local models, use your Azure deployment, or mix whatever works. Our compliance team loves this flexibility - no vendor lock-in means no vendor risk.

I learned this the hard way when GitHub Copilot went down for 6 hours last month and our whole team was dead in the water. With Continue, if OpenAI shits the bed, you just switch to Claude or your local models and keep coding.

Additional Resources:

• Continue Quick Start Guide - Interactive tutorial covering all features
• Complete configuration reference - YAML config documentation
• Model provider setup guides - Step-by-step for different AI providers
• Ollama local model integration - Using Llama 3.1 locally
• Continue Discord community - Active support and troubleshooting
• Continue GitHub Issues - Bug reports and feature requests

The docs say "quick 5-minute setup." That's horseshit. Basic installation? Sure, 10 minutes. Getting it configured so it doesn't drive you insane? Plan on spending a weekend.

Step 1: The Easy Part (5 Minutes)

VS Code: Install the Continue extension from the marketplace. Click, wait, done. The Continue icon appears in your sidebar.

JetBrains: Install from the JetBrains Plugin Marketplace. Works with IntelliJ, PyCharm, WebStorm, etc. Also takes 2 minutes.

CLI (Beta): There's a command-line version if you hate yourself. Needs Node.js. I've literally never seen anyone use this in the wild.

Step 2: Configuration Hell (1-3 Hours)

This is where the fun begins. Continue needs to know which models to use, and the config file is JSON that will make you question your life choices.

Cloud Models - The Expensive Route

First you need API keys. OpenAI costs $0.002 per 1K tokens, which sounds cheap until you get your first $127 bill. Claude costs more but actually thinks before it codes.

Pro tip: Start with OpenAI GPT-4o-mini for testing. It's cheap and works well enough to validate your setup before committing to expensive models.

Local Models - Free but Painful

Ollama setup is "straightforward" if you enjoy pain. Models are 4-7GB downloads that take forever, and you need 16GB+ RAM or your laptop turns into a jet engine. Popular ones are CodeLlama, DeepSeek Coder, and Codestral. Good luck.

Once you get Codestral or DeepSeek working, they're decent. But they're slow as hell compared to cloud models. Great if you can't send code to OpenAI, terrible if you have ADHD. Quantized models are faster but dumber.

Azure OpenAI - Enterprise Tax

If your company has Azure OpenAI, that's your path. More Azure auth bullshit to deal with, but at least compliance won't yell at you.

Step 3: The Fun Stuff (If You're Into That)

Custom Rules and Prompts

Custom rules let you enforce your team's coding standards. Want the AI to always suggest TypeScript interfaces instead of types? Write a rule. Want it to prefer functional components? Another rule.

Real talk: 90% of developers never touch this and just live with whatever the AI spits out. If you're the kind of person who spent 3 hours customizing your shell prompt, you'll love this shit.

MCP Tools - The Killer Feature

This is where Continue gets interesting. MCP tools let Continue talk to other services:

• Create Linear tickets directly from code comments
• Read GitLab issues to understand context
• Pull documentation from Confluence
• Build custom integrations with your internal APIs

Setup time: 30 minutes per tool. Worth it if your team actually uses these services.

Team Deployment - When Shit Gets Real

Continue Hub - Actually Useful Enterprise Dashboard

Continue Hub is one of the few enterprise dashboards that doesn't suck. Set team-wide model defaults, distribute API keys, share custom rules. Your developers can still override settings locally, which prevents the usual "enterprise software is terrible" complaints.

Setup time: 1 hour for basic team config, half a day for advanced rules and integrations.

Security and Compliance

Most security teams ask three questions:

1. "Where does our code go?" - Anywhere you want. Local models keep everything on-premises.
2. "Can we audit the AI interactions?" - Yes, Continue logs everything.
3. "What happens if Continue disappears?" - It's open source, so you can fork it.

Performance Reality Check

• Local models: Slow unless you have a gaming rig with 32GB+ RAM
• Cloud models: Fast but expensive if your team is chatty with the AI
• Mixed setup: Use local for sensitive code, cloud for everything else

The Configuration Trap

Here's what happens: You start with basic OpenAI setup (10 minutes). Then you want local models (2 hours). Then team sharing (1 hour). Then MCP tools (3 hours). Then custom rules (weekend project).

Shit that will ruin your weekend:

• Access to fetch at 'http://localhost:11434' has been blocked by CORS - Ollama CORS is fucked by default
• Rate limit exceeded every 3 minutes during testing because OpenAI's free tier is garbage
• Context length exceeded error when your codebase is bigger than a hello world app
• Request timed out after 30000ms because whoever thought 30 seconds was enough for a 7B model was high
• Continue v0.8.x breaks configs from v0.7.x with zero migration help

Before you know it, you've spent 20 hours configuring an AI assistant. Sometimes the simplest solution is just using Copilot and getting back to coding.

But if you value model flexibility and data privacy, those 20 hours are worth it.

Helpful Resources for Setup:

• Continue installation guide - Official installation steps
• Autocomplete configuration - Deep dive into autocomplete setup
• Model provider comparisons - Choose the right AI models for your needs
• Open WebUI integration - Alternative UI setup
• Continue community guides - Third-party tutorials and tips
• Configuration troubleshooting - Fix common setup issues