Frequently asked questions (FAQ)

Use LEVEL11 as a free local-model coding machine. Accounts, costs, computer setup, your first web page, offline use and troubleshooting.

LEVEL11 + Ollama + a model downloaded to your computer gives you a coding environment without a separate model API bill. LEVEL11 provides the editor, file editing, terminal and agent tools; the model on your computer generates replies and tool calls.

Start with Download LEVEL11 → Install and connect a local model → Build your first web page below.

Jump to: Free usage · Setup and connection · Coding with it · Offline use and data · Troubleshooting

Free usage

1. Can I really use LEVEL11 as a free local-model coding machine?

Yes. Download LEVEL11, run a downloaded local model with Ollama, llama.cpp or MLX, and connect it. These model requests need no paid LEVEL11 plan or model-provider API key, and LEVEL11 does not charge per-token model fees for them.

Free here refers to the app subscription and model API fees for this local setup. You supply the computer, electricity and storage. Paid external APIs, hosting or deployment services can have their own charges. Also check the distribution and usage terms of the model you choose.

2. Can I use it without an account, API key or payment card?

All three are optional for local models. Choose Later on the onboarding sign-in screen. If shown a choice of model connections, choose Local model, or finish the introduction and connect under Settings → Local model.

You can also use a local model while signed in to an existing LEVEL11 account. Check the model selected for your conversation and any model override in its agent preset.

3. Does local coding stop when I use up my Free plan tokens?

LEVEL11 cloud-plan daily and monthly quotas do not apply to local model requests. You can keep using a selected local model after exhausting cloud Free usage, provided its server is running.

Connection	Where the model runs	Credentials and usage
Downloaded local model	Model server on your computer	No LEVEL11 account or API key; separate from cloud-plan quotas
LEVEL11 account cloud model	Model service through the LEVEL11 gateway	Account required; plan allowances and limits apply
Your own API key (BYOK)	Chosen model provider	Provider key required; provider pricing and limits apply

Local models still have memory, processing speed and context limits. See Account and billing for cloud-plan details.

4. Will Auto automatically choose a free local model?

No. Auto in the model picker is the LEVEL11 router, not an automatic local-model selector. Choose Local model explicitly and check for a composer label such as Ollama · qwen3.5:4b.

An agent preset's model override takes precedence. Edit the preset to use local or to follow the session model. See Rules and MCP.

Setup and connection

5. Do I need Ollama, llama.cpp and MLX together?

Only one is needed. Ollama offers an installer and model download commands for a first setup. Use llama.cpp for an existing GGUF file, or MLX for an MLX-format model on Apple Silicon. The backend and URL in LEVEL11 must match your running server.

6. What computer and model should I start with?

LEVEL11 has macOS, Windows and Linux builds. The MLX instructions are for Apple Silicon Macs; on other computers, prepare Ollama or llama.cpp for your operating system.

The first-connection example uses Qwen3.5 4B. Check chat and file access with a small model before changing models to match your free memory and task difficulty. Model-file size is not the entire memory requirement: the operating system, other applications and conversation context also need memory. A particular RAM capacity alone cannot guarantee that a setup fits.

7. What is the shortest setup sequence?

Install and start Ollama, then leave the server running. Set its context length to 8192 for this example.
Download a model and check its name in a new terminal:

ollama pull qwen3.5:4b
ollama list

Under LEVEL11 Settings → Local model, set Backend to Ollama, Server URL to http://127.0.0.1:11434, and Model name to qwen3.5:4b. Do not append /v1 to the URL.
Set Tool calling to Auto and Context length to the same 8192, then apply with Test connection → Use this model → Save.
Choose Open folder, check the composer's model label and send a short greeting. Then ask it to list files to check tool calling too.

Test connection only checks the model list. The local model guide covers installation, server commands for each operating system and a file-creation check. Next time, start Ollama and LEVEL11 again; the same model does not need to be downloaded again.

Coding with it

8. Can a local model edit files and run terminal commands?

Yes. When the model generates valid tool calls, LEVEL11 executes its file-reading, search, editing and command tools. Reliability depends on the model and server's tool support.

Chat: explain, read and review code. It does not edit files or run commands.
Builder: create and edit files, and run commands according to approval and allowlist rules.
Auto-Run: allows more tool execution automatically. Dangerous commands still follow approval rules.

Start with Builder and inspect changed files and requested commands. See Agents for modes, approvals, checkpoints and revert.

9. How do I build my first web page?

Open an empty test folder, choose Local model + Builder, and try this request. A small page without external packages lets you check file creation first:

Create a personal introduction page in index.html and style.css in this folder.
Do not use external packages or CDNs.
Include only a heading, a one-line introduction and a contact section.
When finished, list the changed files and explain how to check the result.

Inspect the files and open index.html in a browser. Follow up with one change at a time, such as Use a dark background. or Adjust the spacing for a phone screen. Use Revert on the relevant turn to restore an unwanted change.

10. Does installing a model replace Node.js or Python?

Your project still needs its runtime and tools. A model server runs the AI; Node.js, Python or a compiler runs the generated program. A static HTML/CSS page opens in a browser, while projects with development servers, tests or packages need those tools and dependencies installed.

Downloading packages for the first time or using an external API can require internet access. Selecting a local model does not automatically prepare every project runtime or dependency.

11. Will it perform like a large cloud coding model?

Equivalent results are not guaranteed. Code quality, tool reliability and speed depend on the model, quantization, memory and context length. With a small model, specify one file, one change and one way to verify it.

Ask it to read relevant files and make a small change before assigning a whole-project rewrite. If needed, try another model within your memory budget. In agent presets, remove unused tools, MCP servers and lengthy instructions to simplify requests.

Offline use and data

12. Can I keep coding without internet access?

With LEVEL11, the model server, model files and project dependencies prepared in advance, local conversation and file work can run offline. Downloads, web tools, external MCP servers, external APIs and remote deployment need network access.

Choose a downloaded local model in Ollama too. An Ollama installation does not make a selected cloud model run offline. The Ollama quickstart distinguishes local and cloud models.

13. Is my code sent to LEVEL11's model server?

Local inference requests go directly to the configured server URL, bypassing the LEVEL11 model gateway. A model running on your PC at 127.0.0.1 processes requests there. Another computer's address or an external server sends requests to that destination instead.

Web tools, MCP servers and integrations communicate with their own services. Selecting a local model does not block every feature's network access.

Troubleshooting

14. Chat works, but the model does not read or edit files

First open a folder and use Builder for edits. Check that the agent preset enables the relevant file tools. If calls still fail, try Settings → Local model → Tool calling → Text, save, and test one small task.

Native tool calling in llama.cpp needs --jinja and a tool-aware chat template. If a model cannot follow the calling format, you may need another model. See tool calling settings.

15. What if replies are slow or cut off?

The first reply can take longer while the model loads. If it stays slow, reduce other applications' memory usage or try a smaller quantized model and a shorter request. Longer context uses more memory.

For cut-off replies or context errors, align the server and LEVEL11 Context length, and check Max output tokens. Increasing the app value alone does not enlarge the server's window. See context settings and the troubleshooting table.

16. Does stopping the server automatically switch to paid cloud models?

A failed local request does not automatically switch to a paid cloud model. Restart the server, check its address, port and model name, and retry. A terminal-launched server stops when you close its window or press Ctrl+C.

If a cloud model is still displayed, check Use this model → Save, the conversation's model selection and the preset's model override. If the issue persists, use the support page with your OS, memory, model name, backend version and error message.