Escape AI: free local AI coding with Ollama — what actually runs on your machine
Escape AI can connect to a local Ollama install so Ask, Plan, and text-based Edit run against a model on your own hardware instead of a cloud API — no per-request cost, and nothing sent off your machine for those requests. Here's exactly what that covers, what still uses a cloud model, and how to turn it on.
What "local" actually means here
Local means the model itself runs on hardware you control — your laptop, workstation, or a server you operate — via Ollama. Escape AI never assumes a fixed model name: it asks your Ollama install which models are actually pulled and lists exactly those, so the picker always reflects what you really have installed, not a hardcoded suggestion.
What runs locally, and what doesn't
Ask mode and Plan mode both work fully against a local model — plain chat and structured multi-step plans don't need function-calling, so a local model handles them the same way a cloud model would. Text-based Edit (asking for a complete rewritten file, without tool calls) also works locally.
Agent mode — the real tool-calling loop that reads files, applies diffs, and runs terminal commands — stays on a cloud model. Reliable function-calling (the model correctly calling a tool instead of just describing what it would do) is inconsistent enough across local and non-OpenAI models that this app doesn't route that specific loop through anything but a proven tool-calling model, regardless of which model you've picked elsewhere in the composer. The app tells you when that's happening rather than silently switching models on you.
Setting it up
1. Install Ollama and pull a model
Install from ollama.com, then pull a coding-capable model, e.g. ollama pull qwen2.5-coder. Any model you've pulled will show up automatically — there's no fixed list to pick from.
2. Point Escape AI at it
By default Escape AI looks for Ollama at http://localhost:11434 (Ollama's own default). If you're self-hosting the Escape AI backend, enable it with the OLLAMA_ENABLED setting; the hosted app surfaces this as a normal option in the model picker once available.
3. Pick it from the model picker
Open the model picker in the chat, desktop, or IDE composer — a "Local" section lists your real, discovered Ollama models. Select one, and Ask/Plan/Edit-text requests route to it.
When to reach for a cloud model instead
- Agent mode's real tool-calling loop — always cloud, by design, regardless of your local-model selection.
- When your local hardware can't comfortably run a large enough model for the task — a small local model will answer faster but reason less well than a frontier cloud model.
- When you're already paying for a cloud plan and prefer its speed/quality trade-off for everyday Ask/Plan use — local is an option, not a requirement.
Try Escape AI free — no card required, and local model support is available once you've got Ollama running.