Mnemo AI¶

A local agentic AI assistant for developers and power users. Mnemo AI runs in your terminal, connects to tools through MCP (Model Context Protocol), can read and edit files, use shell/git safely, search documents with RAG, and remember useful context across sessions.
It supports local and hosted model providers: Ollama, Amazon Bedrock, Bedrock Mantle, OpenAI, Anthropic, Amazon SageMaker AI, and LiteLLM.

Quick start¶
On first run, Mnemo AI launches an interactive setup wizard and writes your user config to:
You only need Python 3.11+ and access to at least one model provider. For the easiest local setup, install Ollama and pull a chat model before running Mnemo AI.
What Mnemo AI can do¶
- Use multiple LLM providers — local Ollama or hosted providers such as Bedrock, OpenAI, Anthropic, SageMaker AI, and LiteLLM.
- Work with your files — read, search, write, and precisely edit text, code, CSV, JSON, PDF, and DOCX files.
- Use developer tools safely — shell execution, git safety checks, todo tracking, plan mode, and background tasks.
- Search and ingest knowledge — RAG over documents, web crawling, and optional Brave web search.
- Learn over time — persistent memory, user profiling, episodic memory, ACE playbook strategies, and authored agent skills.
- Handle multimodal tasks — optional vision model support for image analysis.
Choose your path¶
| Goal | Start here |
|---|---|
| Install and use Mnemo AI from scratch | Getting Started |
| Fix a setup or runtime problem | Troubleshooting |
| Learn the chat commands and feature toggles | Usage |
| Configure providers, models, prompts, RAG, and memory | Configuration |
| Orchestrate complex tasks: query routing, workers, and sub-agents | Orchestration & sub-agents |
| Give the assistant memory: profiles, MEMORY.md, steering, episodic, playbook | Memory & learning |
| Add knowledge: RAG, web search, and web crawling | Knowledge & web tools |
| Connect external MCP servers | External MCP servers |
| Use todos, fast search, precise edits, and background tasks | Everyday tools |
| Control what runs without asking, and what is refused outright | Safety & permissions |
| Look up a tool's exact parameters | Tools reference |
| Find where config, memory, and session state live on disk | The ~/.mnemoai directory |
| Contribute or run tests | Development |
| Understand the internal design | Architecture |
| Browse the detailed per-file map (repo-only, agent/contributor reference) | Architecture Reference |
Typical workflow¶
- Install Mnemo AI.
- Run
mnemoaiand complete the first-run wizard. - Ask a simple question or request a file operation, for example:
- Enable optional features as needed:
- RAG/document search needs an embedding model.
- Web search needs a Brave Search API key.
- Vision needs a vision-capable model.
- External MCP tools are configured in
~/.mnemoai/mcp/mcp.json.
Where configuration lives¶
For normal installs, edit:
You can override this with:
The first run also seeds readable examples under ~/.mnemoai/config/ and ~/.mnemoai/mcp/.