graph TD
A[用户请求] --> B[RAG 搜索请求]
B --> C{"返回结果"}
C -->|"成功"| D[展示详情]
C -->|"失败"| E[显示错误]
F[用户输入] --> G[Chat 模块]
G --> H{"生成回复"}
H -->|"成功"| I[发送回复]
H -->|"失败"| J[处理错误]
Chat · Search your files and the web · Run scripts · Build multi-step AI workflows · Execute AI agent skills - all offline-capable, all on your machine.
⭐ If Askimo saves you time, a star helps others find it - and keeps the project going. Star on GitHub →
You shouldn’t have to choose between the best AI model, your privacy, and getting real work done.
One app, every model. Stop juggling browser tabs. Chat with OpenAI, Claude, Gemini, Grok, or a local Ollama model, switch in seconds, no copy-pasting.
Built as a native desktop app. Not a web wrapper. Starts fast, runs lean, and stays responsive even after hours of use and thousands of messages in a single conversation.
Long conversations that actually work. No crashes, no tab reloads, no lost context. Askimo handles deep, extended sessions the way a real desktop app should.
More than just chat. Delegate real work to autonomous agent CLIs (Claude Code, Codex, Antigravity), chain multi-step AI Plans from a form UI, and connect MCP tools, all from the same app.
Skills, managed once, used everywhere. Define a skill independently of any single agent, then run it with whichever agent CLI fits the job. No per-agent duplication or lock-in.
Talk to it, not just type. Dictate messages and have responses read back to you. Fully local/offline speech-to-text and text-to-speech options included.
🔒 Privacy by design. Your files, RAG index, conversation history, and telemetry all stay on your machine, nothing is uploaded, ever. Local RAG, local SQLite storage, local usage/cost tracking. The only network calls are the ones you configure (your chosen AI provider).
See It in Action
Agents - select the installed agent CLI and delegate a goal:
macOS 11+, Windows 10+, Linux (Ubuntu 20.04+, Debian 11+, Fedora 35+)
Memory
50–300 MB (AI models require additional memory depending on provider)
Disk
250 MB
Features
AI & Providers
Multi-provider - Switch between OpenAI, Claude, Gemini, Grok, Ollama, LM Studio, Docker AI, OpenRouter, NVIDIA NIM, Together AI, vLLM Server, or any OpenAI-compatible endpoint per session
Vision - Attach images to conversations; works with any multimodal model
Search & Data
Web search (multiple backends) - Search the web with DuckDuckGo (no API key), Brave Search API, Tavily, Serply, or your own SearxNG instance
Local RAG - Index local folders, files, and web URLs. Hybrid BM25 + vector retrieval with an AI classifier that skips retrieval when the query doesn’t need it. Your data never leaves your machine.
Workflows & Extensibility
Plans (agentic workflows) - Chain multi-step AI pipelines from a form UI. Each step builds on the previous; progress shown live. Export as PDF or Word. Define your own plans in YAML or generate them by describing your workflow in plain English.
Script runner - Execute Python, Bash, and JavaScript from chat. Python runs in an auto-managed virtualenv with automatic dependency installation.
MCP tool integration - Connect MCP-compatible servers via stdio or HTTP
AI Agents (CLI)
Run autonomous coding agents - Delegate a goal to Claude Code, OpenAI Codex, or Google Antigravity directly from Askimo, with live streamed tool calls, thinking, and status
Reusable skills - Define a skill once and materialize it into any supported agent’s native skill-discovery folder, so it’s just as invocable there as in Askimo’s own chat
Voice
Speech-to-text - Dictate messages via OpenAI’s transcription API or a fully local/offline built-in Whisper engine (whisper.cpp)
Text-to-speech - Have AI responses read aloud via OpenAI’s TTS API or a fully local/offline Piper endpoint, with optional auto-play for hands-free “conversation mode”
Reliability & Privacy
Persistent sessions - Conversations stored in a local SQLite database, restored on restart
Local telemetry - Token usage, cost estimates, RAG performance per provider. Nothing uploaded.
评论