---
title: FAQ
description: Short answers to the questions people ask first.
---
# FAQ

## Does Lampson send my code anywhere?

Only to the model provider you chose, as part of the conversation — the same as pasting into a chat window, but selective: the agent reads the files it needs, not your whole project. Lampson has no server of its own. With Ollama, nothing leaves your machine at all.

## Can it delete my files?

Inside the project, editing and writing files does not ask (that is the job). Recursive deletes, `git reset --hard` and the like are *risky* and stop for your approval in the default mode; the truly destructive commands are refused in every mode. Outside the project folder it cannot reach at all. See [Permissions](/en/docs/permissions).

## Which model is best?

Any capable coding model works. DeepSeek is a good default on price and quality; Anthropic and OpenAI models are excellent; Kimi and Groq are fast. Try two — switching is one command.

## Is it free?

Lampson is open source (MIT). You pay your model provider for what you use; Lampson adds nothing on top. Prompt caching is on where the provider supports it, which keeps long sessions cheap.

## Terminal or web?

Same agent, same sessions. The terminal is faster to start and fits your workflow if you live there. The web UI shows more at once — file tree, live server logs, sub-agents, plugins, schedules — and is where approvals from scheduled tasks show up.

## Why is it written in Synsema?

Because in [Synsema](https://synsema.org) a program declares what it may touch — files, network, processes — and the runtime enforces it. "The agent can only touch this project" is a fact about the code, not a promise. It also made the whole harness short enough to read.

## Does it work offline?

With Ollama, yes.

## Where do I report a problem?

[GitHub issues](https://github.com/kitecosmic/lampson/issues). Your OS, the command you ran and what you saw is plenty; `/trace` output helps.