Verified for v0.4.4 on
Configuration
Settings is divided by purpose. Press Ctrl+Space to search for a settings section, or open Settings from the sidebar.
Models and API keys#
Open AI Models & API Keys to add or edit a model. A provider preset fills in an API base and model ID. Both fields remain editable.
| Field | What it controls |
|---|---|
| Name | Label shown in model selectors |
| Provider | Request formatting and provider-specific handling |
| API base | Complete endpoint used by the Rust backend |
| API key | Credential encrypted in secrets.enc |
| Model ID | Exact identifier sent to the endpoint |
| Supports Image Inputs | Whether the composer accepts images for this model |
| Context size | Context budget used when assembling requests |
| Maximum output tokens | Output space reserved before input context is assembled |
| Temperature | Per-model sampling value from 0.0 to 2.0 |
| Thinking level | Provider-aware reasoning preference, when supported |
| System prompt override | Instructions used for this model instead of the global prompt |
| Allow local network | Permission to contact an exact granted local endpoint |
Changing a preset does not rewrite models you already saved. Select the preset again or edit the model ID when you want the new default.
Connection checks run in the background every five minutes unless background activity is disabled. A failed check is a diagnostic signal; it does not change the saved credential or model ID.
Chat configuration#
Chat Configuration contains four defaults:
- the enabled search provider used by
search_query; - the enabled fetch provider used by
fetch_urland the workspace Browser panel; - global temperature, from 0.0 to 2.0, with a default of 0.7; and
- maximum tool steps, from 1 to 25, with a default of 25.
Per-model temperature takes precedence when it is set on that model.
System prompts#
Open System Prompts to change the default instructions sent with new conversations. The same section controls automatic title generation and its model.
A model can have its own system-prompt override. A project can also have project instructions. When a project contains AGENTS.md at its root, Sythoria reads it for project runs and adds it to the tool context.
Prompts and project instructions consume context and can be sent to the selected hosted provider.
Projects and tools#
Project permissions are configured under Project Workspaces:
| Permission | Available behavior |
|---|---|
| Read | List, search, read, and inspect Git status or diffs |
| Write | Read tools plus approved file writes, edits, and Git commits in an isolated worktree |
| Full | Write tools plus shell commands; each shell command still needs native confirmation |
See Project Workspaces for exclusions, approvals, and Apply/Discard behavior.
Search and fetch#
Sythoria v0.4.4 supports Google Custom Search, SearXNG, and Firecrawl for search. Page fetching supports Firecrawl and Jina Reader.
See Search & Page Fetching for required fields and network rules.
Keyboard shortcuts#
The shortcut editor uses Ctrl notation. On macOS, Sythoria maps application shortcuts to the platform equivalent where supported. Ctrl+Space is a separate fixed global shortcut for settings search.
| Default | Action |
|---|---|
Ctrl+K | Open conversation picker |
Ctrl+L | Focus chat input |
Ctrl+N | New conversation |
Alt+Shift+S | Capture Appshot |
Ctrl+[ / Ctrl+] | Move backward or forward through conversation navigation |
Ctrl+P | Open file picker |
Alt+ArrowUp / Alt+ArrowDown | Select previous or next conversation |
Ctrl+, | Open Settings |
Ctrl+/ | Open the model selector |
Ctrl+B | Toggle the sidebar |
Ctrl+= / Ctrl+- / Ctrl+0 | Zoom in, zoom out, or reset zoom |
ArrowLeft / ArrowRight | Previous or next image in the image viewer |
Escape | Stop generation |
Ctrl+Shift+S | Toggle web search |
Ctrl+Shift+C | Toggle comparison mode |
Ctrl+Shift+R | Retry the last message |
Ctrl+Shift+Backspace | Delete the active conversation |
Ctrl+ArrowDown | Scroll to the bottom |
Ctrl+Shift+V | Toggle voice input |
Ctrl+Shift+T | Toggle theme |
Ctrl+Shift+P | Open command palette |
F2 | Rename active conversation |
Ctrl+E | Export active conversation |
Alt+P | Pin or unpin active conversation |
Ctrl+Shift+W | Open Project Workspaces |
Input limits#
- A message can contain up to 10,000 characters.
- A draft can contain up to six attachments.
- Each attachment can be up to 10 MB.
- Image formats are PNG, JPEG, GIF, and WebP.
- Text attachments include common text, Markdown, data, web, shell, and programming-language files. See Attachments & Vision for the full list.
Local storage#
Encrypted application data is stored in the platform application-data directory:
- Windows:
%APPDATA%\com.sythoria.sythoria-desktop\ - macOS:
~/Library/Application Support/com.sythoria.sythoria-desktop/ - Linux:
~/.local/share/com.sythoria.sythoria-desktop/
Models, projects, network policy, search configuration, MCP configuration, preferences, and credentials use separate encrypted files. Conversations use encrypted snapshots. The root encryption key is kept in the operating system credential vault.
Custom skills are stored separately as plaintext under ~/.agents/skills.
Conversation export writes Markdown to a path you choose. Clear All Chats removes conversation history but keeps configuration. Factory Reset removes conversations, encrypted settings, and Sythoria credential records; it does not remove external backups or exported files.