> For the complete documentation index, see [llms.txt](https://docs.perception.cx/perception/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.perception.cx/perception/enma-ai/getting-started/faq.md).

# FAQ

<details>

<summary>Windows says "unknown publisher" on install</summary>

Expected. The installer isn't code-signed yet. Click "More info", then "Run anyway".

</details>

<details>

<summary>Which endpoint should I use?</summary>

Anything OpenAI-compatible. Free and local: LM Studio or Ollama. Cloud: OpenRouter gives you most models with one key. Enter the base URL ending in `/v1`, then use **Fetch models** to pick a model.

</details>

<details>

<summary>An agent returns empty replies or "reconnecting"</summary>

Usually a gateway or model quirk. Enma retries dropped and empty replies automatically and repairs malformed tool calls. If one agent keeps failing: set its temperature to "model default" (some reasoning models reject pinned temperatures) or give it a fallback chain in Settings.

</details>

<details>

<summary>How do I report a bug?</summary>

Open the chat actions menu and choose **Export debug transcript**. It saves a Markdown file with a low-level table of recent model calls (finish reasons, stream state, where tool arguments cut off, timings) plus the conversation. It contains no API keys and post it with your report.

</details>

<details>

<summary>An update is stuck at 0%</summary>

Restart the app and check again later. If it persists more than a few hours, report it.

</details>

<details>

<summary>Where is my data?</summary>

Config, chats and memory live in `%APPDATA%\Enma`. Your projects are only ever touched inside the folder you picked for the chat. Uninstalling from Windows Settings removes the app; your projects are never deleted.

</details>

<details>

<summary>Can agents push to GitHub or run destructive git?</summary>

No. Agents commit only when asked and never push, rebase or rewrite history, even in full-auto Autopilot. The GitHub buttons in the Git panel are user-only.

</details>

<details>

<summary>Does my code or key go anywhere?</summary>

Only to the endpoint you configured. There's no telemetry backend; the only network calls are your model endpoints, media providers you add keys for, web search when used, and the update check.

</details>

<details>

<summary>A run went wrong. Can I undo it?</summary>

Yes. Per-turn checkpoints let you restore files to any earlier turn (works without git), and restores are themselves undoable. Isolated runs never touched your files in the first place until you merged.

</details>