AI Tools & Environments¶
1. The Tool Landscape¶
AI tools in 2026 span a wide range — from backend model servers to fully autonomous coding agents. Most tools are now hybrid: they can run models locally AND connect to cloud providers.
The ecosystem breaks into 8 categories:
- Model Servers — Backends that run/serve models via API (no UI)
- AI Desktop Apps — Self-contained apps: local runner + chat UI + optional cloud
- AI Web UIs — Browser interfaces that connect to model servers or cloud APIs
- Cloud AI Assistants — Hosted chat interfaces (ChatGPT, Claude.ai, Gemini)
- Cloud Inference Platforms — Run open-source models on rented hardware
- AI IDEs & Coding Assistants — Code editors with integrated AI
- CLI Coding Agents — Terminal-native agentic coding tools
- Autonomous Agents & No-Code — Fully autonomous or no-code AI development
2. Model Servers (Backend — No UI)¶
Pure inference backends that load models and expose an API. Other tools (UIs, IDEs, scripts) connect to them.
| Tool | Interface | Local | Cloud | Platform | Key Feature |
|---|---|---|---|---|---|
| Ollama | CLI + API | ✅ | ✅ (subscription) | Cross-platform | One-line setup, OpenAI-compatible API |
| LocalAI | API | ✅ | ❌ | Cross-platform | Drop-in OpenAI replacement |
| Foundry Local | SDK / API | ✅ | ❌ | Windows | DirectML, .NET SDK |
| Lemonade | CLI + API | ✅ | ❌ | AMD Radeon | Unified runtime: LLMs, ASR, TTS |
| Docker Model Runner | Docker API | ✅ | ❌ | Cross-platform | Containerised model serving |
Key Profiles¶
Ollama — The developer's backbone. One command to pull and run:
Exposes an OpenAI-compatible API at localhost:11434. Powers Open WebUI, AnythingLLM, and many other tools. Supports both local models and cloud frontier models (subscription).
Foundry Local — Microsoft's Windows-native runner. Uses DirectML for GPU acceleration, integrates with .NET and Python SDKs. Pre-bundled models optimised for Windows hardware.
Lemonade — AMD's unified AI runtime for Radeon GPUs. Runs LLMs, speech recognition, and text-to-speech in a single package. Optimised for AMD hardware that's often underserved by the NVIDIA-centric ecosystem.
3. AI Desktop Apps (Runner + Chat UI)¶
Self-contained desktop apps that bundle a local inference engine with a chat interface. Most now also support cloud model APIs for access to frontier models.
| Tool | Local Runner | Cloud APIs | RAG | Cost | Key Feature |
|---|---|---|---|---|---|
| LM Studio | ✅ (llama.cpp) | ✅ (paid tier) | ❌ | Free | Visual model browser, local API server |
| Jan | ✅ | ✅ (paid tier) | ❌ | Free, open-source | Privacy-first, clean UI, hybrid mode |
| GPT4All | ✅ | ❌ | ✅ Built-in | Free, open-source | Simplest setup, bundled document chat |
Key Profiles¶
LM Studio — The friendliest entry point for local AI. Browse and download models through a polished GUI. Can expose a local API server for other tools to connect to. Now supports cloud model subscriptions for frontier models like GPT-4o and Claude.
Jan — Privacy-first with a clean UI. Works fully offline, but can seamlessly switch to cloud APIs when you need frontier model capabilities. Open-source.
GPT4All — The beginner's choice. Download, install, chat. Built-in RAG lets you drag-and-drop documents for Q&A without any setup.
4. AI Web UIs (Frontend — Needs Backend)¶
Browser-based interfaces that connect to model servers or cloud APIs. They don't run models themselves — they need a backend like Ollama or a cloud provider.
| Tool | Connects To | Multi-user | RAG | Cost | Key Feature |
|---|---|---|---|---|---|
| Open WebUI | Ollama, cloud APIs | ✅ | ✅ | Free, open-source | "Self-hosted ChatGPT" |
| AnythingLLM | 35+ providers | ✅ | ✅ Built-in vector DB | Free, open-source | Universal provider hub |
Key Profiles¶
Open WebUI — "ChatGPT, but self-hosted." Browser-based, multi-model, multi-user, with built-in RAG. Connects to Ollama via docker compose up. The go-to UI for teams running local models.
AnythingLLM — The universal connector. Supports 35+ LLM providers across 4 categories:
| Provider Type | Examples |
|---|---|
| Local runners | Ollama, LM Studio, LocalAI, Foundry Local, Lemonade, NVIDIA NIM, Docker Model Runner |
| Cloud AI vendors | OpenAI, Anthropic, Gemini, Mistral, DeepSeek, xAI, Cohere, Moonshot |
| Cloud inference | Groq, Together AI, Fireworks AI, AWS Bedrock, SambaNova, HuggingFace |
| API aggregators | OpenRouter, LiteLLM, APIpie, CometAPI, Generic OpenAI |
Built-in vector database, document embedding, and workspace management. Desktop app and self-hosted server.
5. Cloud AI Assistants¶
The most widely used AI tools in the world. Cloud-hosted, subscription-based, no setup required.
| Tool | Vendor | Free Tier | Paid | Key Differentiator |
|---|---|---|---|---|
| ChatGPT | OpenAI | ✅ | $20/mo | Most popular, GPT-4o / o3 |
| Claude.ai | Anthropic | ✅ | $20/mo | Best reasoning, extended thinking, artifacts |
| Gemini | ✅ | $20/mo | Deepest Google integration, largest free context | |
| Kimi | Moonshot AI | ✅ | Paid tier | Strong reasoning, multilingual |
| Perplexity | Perplexity AI | ✅ | $20/mo | Search-first with citations |
| Mercury | Inception Labs | ✅ | Paid tier | First diffusion LLM, 1000+ tok/s, code + reasoning |
💡 Key insight: For most people, a cloud assistant is their first and primary AI experience. Everything else in this document serves developers and power users who want more control.
6. Cloud Inference Platforms¶
Run open-source models (Llama, Qwen, Mistral, etc.) on their hardware. You pick the model, they provide the compute. Different from cloud assistants because you're renting inference, not using a proprietary product.
| Platform | Hardware | Differentiator |
|---|---|---|
| Groq | Custom LPU | Fastest inference, ultra-low latency |
| Together AI | GPU cluster | Wide model selection, competitive pricing |
| Fireworks AI | GPU cluster | Fastest compound AI, function calling optimised |
| AWS Bedrock | AWS infrastructure | Enterprise, private VPC, managed |
| SambaNova | Custom SN40L chip | High throughput, enterprise |
| HuggingFace Inference | GPU cluster | Largest model hub, community |
API Aggregators¶
One API → many providers. These don't run models — they route your request to the best/cheapest provider:
| Tool | Value Prop |
|---|---|
| OpenRouter | Unified API, price comparison across providers |
| LiteLLM | OpenAI-compatible proxy, self-hostable |
7. AI IDEs & Coding Assistants¶
Full code editors with integrated AI — the primary way developers interact with AI in 2026.
| Tool | Type | Price | Key Differentiator |
|---|---|---|---|
| Cursor | AI IDE (VS Code fork) | $20/mo | Composer (multi-file), Background Agents |
| Windsurf | AI IDE (VS Code fork) | $15/mo | Cascade (agent manager), Arena Mode |
| Antigravity | AI IDE | — | Google DeepMind, advanced agentic coding |
| GitHub Copilot | IDE Plugin | $10/mo | Enterprise standard, best test generation |
| Zed | Native Editor | Free/usage | Rust-native, starts in <1s, real-time collab |
| Continue | IDE Extension | Free | Open-source, multi-provider + local via Ollama |
| JetBrains AI | IDE Plugin | $10/mo | IntelliJ, PyCharm, WebStorm integration |
Key Profiles¶
Cursor — The most popular AI IDE in 2026. A VS Code fork with deep AI integration: Composer for multi-file edits, tab completion, chat, and Background Agents that run autonomously. The "default choice" for professional developers.
Windsurf — Strong competitor with unique features: Cascade (an agent that manages the full edit lifecycle), Arena Mode (A/B test two models on the same task). More beginner-friendly than Cursor.
Antigravity — Google DeepMind's AI IDE with advanced agentic capabilities. Deep integration with Google's ecosystem and Gemini models. Browser automation, MCP support, and sophisticated planning workflows.
GitHub Copilot — The enterprise standard. Deepest GitHub integration, best test generation, conservative approach to code changes. The safest choice for corporate environments.
💡 Pro tip (2026): Most senior devs use two tools — an AI IDE (Cursor, Windsurf, or Antigravity) for daily editing + a CLI agent (Claude Code or Gemini CLI) for complex architectural tasks.
8. CLI Coding Agents¶
Terminal-native tools for agentic coding — AI that reads your codebase, edits files, runs commands, and iterates autonomously.
| Tool | Vendor | Key Feature | Model Support |
|---|---|---|---|
| Claude Code | Anthropic | 1M context, filesystem + terminal access | Claude models |
| Gemini CLI | Google ecosystem, Gemini models | Gemini models | |
| OpenAI Codex | OpenAI | Best review UI, worktree support | GPT models |
| Aider | Open-source | Git-aware, auto-commits, diff editing | Multi-provider |
| Cline | Open-source | VS Code extension, browser automation, MCP | Multi-provider |
Key Profiles¶
Claude Code — A terminal-native AI agent that goes far beyond autocomplete. It reads your codebase → plans changes → edits files → runs tests → iterates. Full system access. Best for complex refactoring and multi-file feature development.
Gemini CLI — Google's terminal agent, powered by Gemini models. Deep integration with Google Cloud and the broader Google developer ecosystem.
Aider — The git-native AI coding tool. Automatically creates git commits for each change, shows diffs before applying. Supports multiple AI providers including local models via Ollama.
Cline — An in-editor agent (VS Code extension) that creates a plan, then executes file edits, terminal commands, and browser actions. Shows each step for human approval. 5M+ installs.
9. Autonomous Agents & No-Code Builders¶
Autonomous Development Agents¶
Tools that operate with minimal human guidance — give them a task and they plan, code, test, and deploy:
| Tool | Type | Price | How It Works |
|---|---|---|---|
| Devin AI | Autonomous agent | $20/mo | End-to-end: plan → code → test → deploy |
| Replit Agent | Browser IDE | $20/mo | Full-stack from prompt, zero local setup |
No-Code / Low-Code Builders¶
For rapid prototyping without writing code:
| Tool | Best For | Output |
|---|---|---|
| Lovable | Web app prototypes | Deployed React apps from descriptions |
| Bolt.new | Full-stack apps | Working apps in the browser |
| v0 (Vercel) | UI components | React/Next.js components from prompts |
10. Tool Selection Guide¶
What do you need?
│
├── Chat with AI (no setup)?
│ └── ChatGPT, Claude.ai, or Gemini
│
├── Run models locally?
│ ├── Prefer a GUI? → LM Studio or Jan
│ ├── Prefer terminal / API? → Ollama
│ ├── Windows + DirectML? → Foundry Local
│ ├── AMD GPU? → Lemonade
│ └── Browser-based (team)? → Open WebUI + Ollama
│
├── Run open-source models in the cloud?
│ ├── Fastest inference? → Groq
│ ├── Widest selection? → Together AI
│ └── Enterprise / AWS? → Bedrock
│
├── Connect to many providers?
│ └── AnythingLLM (35+ providers, built-in RAG)
│
├── AI-assisted coding?
│ ├── Full IDE? → Cursor, Windsurf, or Antigravity
│ ├── Enterprise / GitHub? → GitHub Copilot
│ ├── JetBrains user? → JetBrains AI
│ ├── Open-source / local? → Continue
│ ├── Terminal agentic? → Claude Code or Gemini CLI
│ └── Git-aware? → Aider
│
├── Fully autonomous development?
│ └── Devin AI or Replit Agent
│
└── Rapid prototype (no code)?
└── Lovable, Bolt.new, or v0