Aider clicked for me once I realised it commits each change as its own git commit, so the model's work is reviewable and revertable like any teammate's. Being model-agnostic means I can point it at whatever is cheapest or strongest that week. It rewards discipline: feed it the right files with /add or it burns tokens guessing. For terminal-first, git-native AI coding it is hard to beat.

Continue earns its place by being the open one you can actually bend: point it at a local model, wire in custom context providers, keep your code on your machine. That flexibility is also the cost of entry, since the config takes real time to get right and the defaults are not magic. Once tuned with a local model it is a solid, private alternative to the closed assistants.

Claude Code is the one agent CLI I trust to change a dozen files and not lose the plot. Living in the terminal means it slots into my existing git and tooling instead of fighting it. It will over-edit if you hand it a vague instruction, so I keep tasks scoped and read every diff, and token cost on long sessions is real. For actual multi-step work though it is a clear step above the autocomplete-style tools.

Cursor is the editor I keep going back to because the multi-file edits actually hold context across a repo instead of just the open file. Tab completion guessing my next edit a few lines down still surprises me. Two honest gripes: on a large monorepo the indexing lags and it sometimes pulls stale context, and the usage-based pricing climbs fast once you lean on the bigger models. Still the smoothest agentic editing I have used.

Ollama is what I reach for when I want a model running locally in under a minute, and that matters more than people admit. The one-line pulls and the OpenAI-compatible endpoint mean I can drop it under existing code with almost no changes. Where it still bites me: concurrent requests get serialized in ways that aren't obvious until you load-test, and juggling VRAM across a few models gets fiddly. For local dev and prototyping though, nothing else is this frictionless.

Open WebUI is the self-hosted chat front-end I actually kept running. Point it at Ollama and you get a clean multi-model UI, RAG over your own docs, and per-user access without sending anything to a cloud. Setup beyond the basic compose takes some patience, and it gets heavy with several models loaded at once. For a private, offline-capable ChatGPT-style UI it is the best I have tried.

LM Studio is what I hand people who want local models without touching a terminal. The model browser and the GGUF quant picker make it obvious which version fits your VRAM, which genuinely lowers the barrier. It is GUI-first though, so for anything scriptable I still drop to a headless runtime, and it is heavier on resources than a bare server. As a local playground and for trying quants, it is excellent.