Parallel branches without losing context
Fork from any message to compare models, prompts, or reasoning paths.
Next-generation AI workspace
Move AI beyond chat windows into a long-running workspace. Mirdel unifies conversations, knowledge bases, notes, translation, image and video work, local models, and extensible workflows so repeated processes become reusable capabilities.
Local-firstSQLite, vector indexes, and sensitive configuration stay local whenever possible
Multi-modelCloud models, compatible endpoints, and a built-in local model runtime
ExtensibleApplets, Skills, and MCP turn workflows into durable capabilities
Workspace
Not another chat client, but a complete desktop AI workspace.
Mirdel breaks daily AI work into clear workbench modules while preserving shared context, search, model configuration, and a local data layer.
Chat and session map
Branch conversations, temporary sessions, quick messages, session notes, and visual relationship maps keep long-running exploration organized.
Knowledge base and local RAG
Import files, folders, and webpages, build local indexes with SQLite and sqlite-vec, and test retrieval quality.
Notes and AI assistant
Edit in Tiptap or Markdown, review AI edits as diff proposals, and commit changes only after confirmation.
Translation and document processing
Handle text, full documents, and dictionary-style parsing with independently configurable translation models.
Image workbench
Unify text-to-image, image-to-image, editing, outpainting, super-resolution, colorization, and style transfer tasks.
Video workbench
Configure provider-specific resolution, duration, frame rate, reference images, and model capabilities for video workflows.
Capabilities
AI infrastructure designed for long-term use.
From branched sessions to local RAG, from model capability configuration to tool-call approval, Mirdel focuses on order and control after an AI workspace starts running for real.
Scenario templates
Package models, parameters, system prompts, knowledge bases, Skills, and MCP policies into reusable presets.
Built-in web search
Mirdel includes SearXNG for aggregated search and supports custom search services.
Layered memory
Current context, session state, cross-session memory, and long-term user preferences work together for sustained work.
Local model runtime
An embedded llama-server can download, load, and unload local models while exposing a local compatible endpoint.
Human approval and encryption
Sensitive Shell and MCP calls require confirmation, while API keys and other sensitive fields are stored encrypted.
Extensible by design
Turn prompts, tools, and workflows into reusable capabilities.
Mirdel's extension system is not just a plugin list. It organizes procedural knowledge, external tools, and productized workflows into clear layers.
Turn stable workflows into clickable mini apps
Applets have their own forms, parameters, interaction state, result views, and separate windows, turning frequent tasks from repeated prompting into launch-and-use workflows.
Inject procedural knowledge into agents
Compatible with SKILL.md organization, with built-in system skills for docx, pdf, pptx, xlsx, skill creation, and more.
Connect the file system, Shell, and external tools
Supports stdio, Streamable HTTP, and SSE, with service lifecycle controls, capability inspection, logs, and per-message policy control.
Local-first
Your data, models, and workflows should not be locked into one service.
Sessions, notes, knowledge bases, configuration, and indexes are stored in a local SQLite database. API keys and sensitive fields are encrypted. You can connect mainstream cloud models, compatible endpoints, or local models.
Unified model and tool entry point
OpenAIAnthropicGooglexAIDeepSeekTongyiDoubaoZhipuMinimaxLocal llama-serverOpenAIAnthropicGooglexAIDeepSeekTongyiDoubaoZhipuMinimaxLocal llama-server
Compatible with OpenAI / Anthropic / Google endpointsBuilt-in SearXNG and document parsingLocal import, export, and secure backups
Download
Start building your desktop AI workspace.
Mirdel supports macOS and Windows. After downloading, configure a cloud model or use the local model runtime to start organizing long-term AI workflows.