Dulus - an open-source CLI AI agent that drives multi-model tool calls.

What is Dulus?

Dulus is an open-source CLI AI agent consisting of approximately 12,000 lines of Python code, supporting over 40 models including Claude, GPT, Gemini, DeepSeek, Kimi, and Qwen. The tool can directly capture visitor sessions from web-based AI applications in browsers and convert them into tool call proxies with 27 functions such as file reading and writing, code editing, Bash execution, and web search. Dulus supports MCP, hot plugin reloading, sub-proxies, offline voice, persistent memory, and automatic snapshots, and provides entry points across multiple platforms including PyQt6 GUI, WebChat, and Telegram.

Dulus's main functions

• Zero-key web session captureAutomatically hijacks browser tabs such as Gemini, Claude.ai, Kimi.com, and DeepSeek, converting web page conversations into a local agent that can be invoked by tools.
• 27 built-in tools: Covers Read, Write, Edit, Bash, Glob, Grep, WebFetch, WebSearch, NotebookEdit, GetDiagnostics, Memory, Tasks, Agents, Skills, etc.
• Multi-provider model switchingSupports Anthropic, OpenAI, Google, DeepSeek, Kimi, Qwen, Zhipu, MiniMax, Ollama, LM Studio, and custom endpoints. (During the session...) /model Real-time switching.
• MCP and Plugin Ecosystem:place .mcp.json It can register any MCP server; the Auto-Adapter can hot-load any Python repository as a plugin with zero configuration.
• Sub-agents and Task ManagementGenerates sub-agents of types such as coder, reviewer, and researcher in an independent git worktree, enabling collaboration through message passing and featuring a built-in task board.
• Offline voice and wake-upOffline voice input is achieved using Whisper-cpp, and voice output is achieved using Kokoro TTS; custom offline wake words such as "hey dulus" are supported.
• Persistent memory and snapshotsDual-range memory (user-level + project-level) sorted by confidence level and relevance; automatically creates checkpoints in each round, supporting one-click rollback of conversation and file status.
• Multimodal interactive interfaceIt offers four interaction methods: REPL terminal, Flask WebChat (accessible on LAN), PyQt6 native desktop GUI, and Telegram Bot.
• Developer Enhancement ModeSSJ Mode provides debugging capabilities such as raw token counting, stream latency timing, tool call inspector, and Prompt injection viewer.

How to use Dulus

• Install:implement pip install dulus or curl -fsSL https://raw.githubusercontent.com/KevRojo/Dulus/main/install.sh | bash One-click installation.
• Configuration key: Set environment variables for any provider, such as export ANTHROPIC_API_KEY=sk-ant-...Alternatively, you can skip using the Ollama native model and browser hijacking mode.
• Start interactionTerminal input dulus Enter REPL, use /model nvidia-web/deepseek-r1 or /model ollama/qwen2.5-coder Choose a model.
• Enable zero-key modeUpon first run, follow the wizard prompts to open the Gemini browser visitor page. Dulus will automatically capture the session, and tool calls can begin without an API key.
• Execute the taskSimply input natural language commands such as "refactor the auth module", and the Agent will automatically read the file, edit the code, run the test, and submit the results.
• Scalability:pass /plugin install 仓库名@URL Hot-load any Python utility library, or /mcp Register external MCP server extension functionality.

Dulus's core advantages

• True zero-cost start-upNo API Key, credit card, or login required. Obtain the complete tool call agent within 30 seconds through browser guest mode, minimizing the barrier to entry.
• Model neutrality and real-time switchingNot bound to a single provider, within the same session /model It allows seamless switching between models such as Claude, DeepSeek, and Kimi, and supports automatic degradation chains.
• Hot-swap ecosystem expansionAuto-Adapter turns any Python repository into a plugin in seconds. MCP and Composio provide 800+ skills, and ecosystem expansion can be achieved without waiting for official adaptation.
• End-to-end offline capabilityIt supports voice input/output, wake word, OCR, and Ollam local models, all of which can run in a completely offline environment, adapting to secure networks and offline scenarios.
• Unified Core for Multiple InterfacesTerminal, Web, desktop GUI, and Telegram share the same Agent core and memory state, allowing users to freely switch between interaction surfaces according to the scenario.

Dulus project address

• Project official websitehttps://dulus.ai/
• GitHub repositoryhttps://github.com/KevRojo/Dulus

Comparison of Dulus's similar competing products

Dulus application scenarios

• Personal Developer Coding AssistantIt replaces traditional IDE plugins, directly driving code reading, writing, refactoring, testing, and submission through natural language, and supports multi-model collaborative review.
• Zero-budget AI experimentsStudents or beginners can experience the advanced tool calling capabilities of cutting-edge models for free using Gemini's guest mode, without needing to subscribe to the paid API.
• Offline/Secure Environment AutomationA fully private agent workflow based on the Ollama local model and offline voice in government, military, or enterprise intranets.
• Telegram remote maintenanceSend commands to Dulus at home or on a server via Telegram on your mobile phone to perform remote operations such as file management and log checking.
• Multi-Agent Collaborative DevelopmentIt distributes code writing, review, and testing to different sub-agents for parallel processing, making it suitable for distributed development of large modules.