deepchat: Local-First AI Agent Desktop Client with Electron

Overview

DeepChat is a local-first AI agent desktop client built with Electron and Vue, available for Windows, macOS, and Linux. It brings model access, tools, agent runtimes, and long-running sessions together, so people can work with cloud APIs or local Ollama models without switching between separate clients.

Its Tape-based approach keeps session history, tool calls, requests, and results recoverable and inspectable. DeepChat also connects external ACP agents and supports controlling sessions through messaging apps.

Key Features

• Manage cloud and local Ollama models in one desktop app, with custom providers that use OpenAI, Gemini, or Anthropic API formats.
• Keep agent sessions recoverable with Tape history and inspect request traces, model metadata, and token budgets.
• Install and enable reusable Skills for specific conversations, with import and export support for compatible tools.
• Connect ACP-compatible agents as selectable entries in the model selector.
• Configure MCP tools, resources, and prompts over StreamableHTTP, SSE, or stdio transports.
• Control sessions remotely through Telegram, Feishu/Lark, QQBot, Discord, and WeChat iLink.
• Work across multiple windows, tabs, and conversations, with Markdown, code, image, and Mermaid rendering.

Use Cases

• Developers can use cloud or local models for coding help, then connect ACP coding agents when they need a separate agent runtime.
• Teams with repeatable workflows can install Skills for tasks such as code review, document work, or PDF processing and enable them per conversation.
• People running long tasks can use Tape history and traces to inspect tool activity and resume sessions.
• Users who want to check on or manage desktop sessions away from their computer can connect a supported messaging app for remote control.

Getting Started

For the desktop app, download the installer for your platform from GitHub Releases. For development, install dependencies and start the app:

pnpm install
pnpm run installRuntime
pnpm run dev

See the README for setup, provider configuration, and build details.

Alternatives

• holaOS: holaOS centers agents alongside existing apps and integrations with shared local memory, while DeepChat focuses on model providers, agent tools, and protocols.
• Kun: Kun focuses on shared agent tasks and approvals across desktop and terminal workflows, rather than DeepChat’s provider- and integration-centered client.
• maka: Maka emphasizes agent execution history in an append-only log and connecting your own model, while DeepChat offers a broader set of providers and integrations.

Considerations

• The latest listed release is a prerelease, so teams adopting it should account for beta-stage changes despite frequent commits and releases.
• Development requires Node.js >=24.21.0 <25; the project does not list Docker as a runtime option.
• The client stores data locally and connects to configured model services. Review provider and messaging integrations against your organization’s data-handling requirements.
• Apache-2.0 permits commercial use, subject to the license’s terms and notices.