Cobolt
@platinum-hill
About Cobolt
This is a cross-platform desktop application that allows you to chat with locally hosted LLMs and enjoy features like MCP support
Overview
What is Cobolt?
Cobolt is a cross-platform desktop application that runs locally on your device, using local AI models via Ollama to provide answers and perform actions on your data without sending it to external servers. It supports the Model Context Protocol (MCP) to connect to tools and data sources.
How to use Cobolt?
Install the appropriate binary for your operating system (macOS, Windows, or Linux) from the releases page. Ensure required system dependencies are installed (Homebrew on macOS, Winget on Windows, or run the linux_deps.sh script on Linux). Then download and configure Ollama models (default is llama3.2:3b for inference, nomic-embed-text for embedding) and add MCP integrations through the app’s JSON configuration file, using the same format as Claude Desktop.
Key features of Cobolt
- Local AI models via Ollama, data never leaves your device
- Model Context Protocol (MCP) integration for external tools and data
- Native memory support for personalized, context-aware responses
- Cross-platform support (macOS, Windows, Linux)
- Open source under Apache 2.0 License
- Customizable model selection for inference, tool use, and embedding
Use cases of Cobolt
- Get answers and perform actions based on your local data without uploading to cloud
- Connect to favorite tools and data sources using MCP servers
- Use a locally running AI that remembers important details about you for personalized responses
- Choose from various open-source Ollama models for different tasks
FAQ from Cobolt
What models does Cobolt support?
Cobolt uses Ollama and supports any Ollama model that supports tool calls. Default inference model is llama3.2:3b, default embedding model is nomic-embed-text. You can change models via app settings or by editing the config JSON file.
Is my data sent to external servers?
No. Cobolt only stores data on your device and uses locally running AI models. It does not send your data or queries to any external server.
How do I add MCP integrations?
Add new MCP servers through the application, which directs you to a JSON file using the same format as Claude Desktop. After adding servers, restart the application or reload integrations.
What platforms are supported?
Cobolt supports macOS (DMG), Windows (EXE), and Linux (AppImage). Downloads are available on the releases page.
Is Cobolt free?
Yes, the project is open source under the Apache 2.0 License. It is currently in early release (v0.0.6) and may be unstable and change significantly.
Basic information
More Desktop Chat MCP clients
🧠 Ultimate MCP Client
dicklesworthstoneDTC MCP
rafaelsztutmanContext-optimized MCP server for DTC e-commerce brands. Connect Claude (or any MCP client) to your Klaviyo and Shopify data with 16 pre-built analytics tools.
Model Context Protocol for Raycast
evanzhoudevKILN-AI
Kiln-AIKiln is a free tool for building production-ready AI systems. It supports RAG pipelines, evaluations, agents, MCP tool-calling, synthetic data generation, and fine-tuning.
Witsy
nbonamyWitsy: desktop AI assistant / universal MCP client
Comments