FAQ
Short answers to the most common LEX AI questions: getting started, choosing among the ten modes, supported coding tools, model choice, git behaviour, reading data files, resuming, and updating.
FAQ
Quick answers to the questions that come up most. The full FAQ ships with the product: run lex ai-faq, which renders the page locally from the installed package and opens it in your browser. Nothing is fetched from the internet, so the answers always match the version you actually have.
How do I start?
Run lex setup-with-ai in your project, open lex ai-dashboard, pick the mode that matches your goal, then start a fresh assistant request describing what you want. The command is hyphenated — lex setup_with_ai does not exist and fails with No such command. See Setup and dashboard.
Which mode should I use?
Choose by intent: interview me and write the requirements down → Brief; new app → Forward; smallest working version → MVP Generator; grow an MVP into the real thing → MVP Completion; change an existing app → Edit; audit without changing anything → Review; write and prove a test suite → Test; the input file format changed → Input; get it running somewhere → Deploy; document existing code → Backward. See Choose the right path.
I do not know what I want yet. Where do I start?
Brief. It interviews you one question at a time and writes .lex/contract.md, then names the build mode to hand off to. It exists because a form of twenty expert-level boxes asks you to author a specification unaided, and nobody does. See Brief.
Why can only one mode's tools appear at a time?
The unified server mounts all ten modes but exposes one surface, so every workflow stays deterministic and the assistant cannot reach for a tool from a process it is not running. Switching flips the surface and refreshes the tool list. See Tool surfaces.
The assistant says a tool is not available — what now?
That is usually a cached tool list after a mode switch (stale_tool_call: true). Refresh the tool list, switch mode if a suggested_mode is named, then retry the same tool. Do not work around it by calling something else.
Which coding tools are supported?
Seven: GitHub Copilot in JetBrains IDEs, in VS Code and in the CLI, plus Cursor, Claude Code, OpenAI Codex and Windsurf. Each gets the agents, slash commands and workspace rules in its own native format. See Supported coding tools.
Does it matter which model I use?
Yes, materially. lex ai-faq renders the current LEX Compatibility Score table, and Model compatibility explains how it is scored and what each tier means in practice.
Does LEX AI push to my GitHub?
Only Forward and MVP Completion: they open a run branch, commit and push per step, and open and merge a PR at the end. Edit opens a run branch, commits each task, pushes and opens a pull request — but never merges. Brief, Backward, Review, Test, Input and Deploy never push. MVP Generator makes local commits only. See Git and GitHub.
Can the assistant read my data file?
Yes. read_input_file reads .csv, .tsv, .txt, .xlsx, .xlsm, .xls and .pdf server-side and returns markdown, and render_pdf_pages rasterises PDF pages so a multimodal model can read a scan or a chart with no text layer. Both are on every mode's surface. See Files and questions.
Can I change the plan or add data mid-run?
Yes, in Forward and MVP Completion. Editing a planning document or dropping in a new data file between steps triggers automatic re-execution from the earliest affected step. Docs drive code, and code drives docs.
Can I stop and resume later, or hand off to someone else?
Yes. State lives in files — the per-mode session manifest, plans/, technical-map/, the run reports — and in git history, not in the chat. Start a new session and call the mode's resume path.
Why was my MVP scope rejected?
You hit a hard cap: 10 requirements of 240 characters each, 5 entities, or 8 fields per entity. Trim the scope and expand later with MVP Completion.
Should I run git myself?
Not in Forward or MVP Completion — the server owns git there, and switching branches mid-run misplaces commits the manifest is still tracking. In the flat modes you commit the output yourself once you are satisfied.
How do I update, and how do I know it worked?
lex ai-update, then lex setup-with-ai to refresh the payload, then lex ai-verify to confirm the on-disk assets match what is installed. Start a fresh assistant request afterwards.
Something is broken and I need help.
Run lex ai-verify first. If you still need support, lex ai-issue-report builds a diagnostic ZIP under .lex-ai-reports/ with credential values masked before anything is written. See Troubleshooting.
Was this helpful?
Your feedback shapes what we write next.
