Sakrylle Chat
Sakrylle Chat is Sakrylle's LLM chat client. The current public release only supports macOS Apple Silicon. It includes built-in Sakrylle API sign-in.
- Support: support@sakrylle.com
Overview
What it is for
- Sign in with a Sakrylle account and start chatting without manually entering an API key
- Manage models, assistants, and conversations in one client
- Use MCP, built-in tools, web search, translation, OCR, TTS, backup, and desktop workflows
Quick Start
Install
| Platform | Architecture | Download |
|---|---|---|
| macOS | arm64 | Sakrylle_Chat_macOS-1.1.15.dmg |
Sign in
Sakrylle Chat requires OIDC sign-in. There is no skip option. After sign-in, the built-in Sakrylle API provider is configured automatically.
| Parameter | Value |
|---|---|
| Issuer | https://oidc1.sakrylle.com |
| Client ID | sakrylle-chat |
| Flow | Authorization Code + PKCE (S256) |
| Scopes | openid profile email models:read chat.completions:create responses:create account:read offline_access |
| macOS callback | sakrylle-chat://oauth/callback |
Tokens are encrypted in flutter_secure_storage, refreshed automatically before expiry, and cleared on logout.
If you see "Unauthorized" after sign-in, your account is usually not activated yet or has a billing-state problem. Check the console at https://ai1.sakrylle.com first, then contact support@sakrylle.com if needed.
First setup
- Sign in and let the app sync models and balance.
- Set default chat, title, summary, and suggestion models on the model page.
- Enable tool-related options per model as needed.
Feature Details
Chat and history
Basic chat
- Responses stream by default
- Supports regenerate, edit-and-resend, branching, and multi-version assistant replies
- Supports multi-select export to Markdown, TXT, or image
- Supports selective text copy inside messages
Temporary chat
- Temporary chats do not appear in history
- They are deleted completely after you leave
- Empty temporary chats show a dedicated empty state
- Temporary mode can be toggled in the chat view
Conversation management
- Create, delete, rename, and pin conversations
- Truncate conversation context
- Auto-generate titles with an LLM
- Auto-generate conversation summaries with an LLM
- Optionally start a new conversation when switching assistants
Global search
- Entry: the search box at the top of the chat sidebar
- Search results can jump back into the matching conversation or message
- History access and global search share the same top area
Assistants
Assistants have independent settings and can override global model behavior.
Assistant basics
| Parameter | Description |
|---|---|
| Name / avatar | Assistant display info |
| Independent model | Overrides the global chat model |
| System Prompt | System instruction |
| Message template | Supports |
| Temperature / Top P | Sampling controls |
| Max Tokens | Max output size |
| Thinking Budget | Reasoning budget control |
| Context message count | Limit for history messages sent upstream |
| Streaming | Per-assistant toggle |
| Chat background | Color or image |
Assistant enhancements
- Persistent memory
- Recent chat references
- Tagging
- Preset messages
- Regex preprocessing rules
- Quick Phrases
- World Book keyword knowledge injection
- Instruction Injection with groups
Models and suggestions
Model selector
- The top model picker shows models by group
- Frequently used models can be pinned
- Switching assistants can also switch models
Model assignments
Sakrylle Chat supports separate models for different tasks:
| Function | Description |
|---|---|
| Chat Model | Primary conversation model |
| Title Summary Model | Auto-title model |
| Summary Model | Conversation summary and recent-chat summary model |
| Translation Model | Translation model |
| OCR Model | OCR model |
| Suggestion Model | Follow-up suggestion model |
| Compression Model | Context compression model |
Chat suggestions
- Suggestions are not always on by default
- Suggestion bubbles are generated only after a Suggestion Model is configured
- The app can also be set to insert a suggestion without sending it immediately
MCP
Sakrylle Chat supports MCP (Model Context Protocol) servers and can bind them per assistant or per conversation.
Transport methods
| Transport | Status | Notes |
|---|---|---|
| Streamable HTTP | Implemented | One of the main remote MCP options |
| SSE | Implemented | Remote compatibility option |
| STDIO | Implemented | For local processes |
STDIO fields
STDIO servers support these fields:
| Field | Description |
|---|---|
command | Launch command |
args | Launch argument array |
env | Environment variables |
workingDirectory | Working directory |
Other MCP behavior
- Tool approval is supported
- Built-in
kelivo_fetchMCP tool is included - Exported configs use
streamableHttp/sse/stdio
Tools and search
Built-in tools are not fixed default capabilities. Availability depends on model capability, whether the model override enables the tool, and the current Sakrylle API model route.
Built-in tool conditions
| Tool | When it is available |
|---|---|
search | Requires the current model override to enable it, and the current model route must support it |
code_interpreter | Requires support from the current model route |
image_generation | Requires support from the current model route |
Search limits
- The current model override must enable
search - The current model route must support built-in search
- Availability still depends on the specific model
Search services
The app also has a standalone web-search settings page for connectivity checks, global enable/disable control, and result-count settings. That is a separate layer from the built-in search tool above.
Multimodal and rendering
Multimodal input
- Image upload and cropping
- Independent OCR model and prompt
- PDF, Word, and text file upload
- Drag-and-drop upload
Rendering
- Markdown
- LaTeX math
- Mermaid
- PlantUML
- Reasoning display with collapse controls
WebView / HTML preview
- The WebView entry lives in the message menu
- It renders the current message into HTML and previews that content
- Preview uses an HTML preview dialog
- Console Logs are available
- This is not a general browser surface for public pages such as the console or docs site
TTS and backup
TTS
- System TTS is supported
- Auto-play, chunked playback, timeline, and selected-text playback are supported
Translation
- Independent translation model and prompt
- A separate translation tab is available
Backup
- WebDAV backup / restore
- S3-compatible backup / restore
- ChatBox / Cherry Studio import
- Storage cleanup by category
Configuration Reference
Key settings
| Setting | Where | What it controls |
|---|---|---|
| Default model set | Model page | Chat, title, summary, suggestion, OCR, translation, and compression models |
| Suggestion Model | Model page | Suggestions stay disabled until this is configured |
| MCP servers | MCP page | Streamable HTTP / SSE / STDIO |
| Search services | Search page | Search engines, connectivity, result count |
Errors and limits
Confirmed limits
- Sign-in is required; there is no guest mode
- Chat suggestions require an explicitly configured Suggestion Model
- Temporary chats are deleted after you leave them
- Built-in
searchis constrained by model, model override, and current model route conditions - Current builds do not expose a visible Sponsor settings entry; rely on the settings items visible in the UI.