Description
Argentum - AG064 is a modular, self-hosted AI agent harness developed in Rust, offering both a command-line interface (CLI) and a native Slint desktop client. The project prioritizes user control and privacy, explicitly stating no subscriptions, telemetry, or cloud lock-in. This design ensures that users retain full ownership and management of their AI agents and data.
Read More
The architecture of Argentum is highly modular, organized into distinct Rust crates covering runtime, tools, policy, storage, and UI components. This structure facilitates maintainability and extensibility, allowing developers to integrate and manage various AI functionalities efficiently. The system supports workspace-scoped provider profiles, including popular language model services like LM Studio, OpenAI, and Anthropic, enabling flexible integration with different AI backends.
Key features include policy-backed execution profiles, such as "Read Only" and "Confirm Before Changes," which provide explicit control over agent actions and data modifications. This focus on explicit execution policies enhances security and user confidence, ensuring that AI agents operate within defined boundaries. The platform also maintains a bounded session trajectory, tracking tasks, tool calls, approvals, and usage, offering transparency into agent activities.
Argentum is released under the Apache-2.0 license, reinforcing its commitment to open-source principles and community collaboration. The project's development involved a complete rewrite in native Rust from an earlier TypeScript implementation, demonstrating a dedication to performance and native integration. The developer, Aleksey Aleksandrovich, built the entire agent harness, including the CLI, native desktop client, and the boundaries around tools, providers, and execution policies.
This tool is ideal for developers and IT specialists who require a robust, self-hosted solution for building and managing AI agents with a strong emphasis on modularity, security, and explicit control. It caters to those who prefer open-source alternatives over SaaS models, avoiding vendor lock-in and ensuring data sovereignty.
- CategoryAI Agent Builders
- Pricing ModelOpen Source
- DeploymentSelf-Hosted
- PlatformsDesktop, CLI
Argentum's Core Features
Modular AI agent harness in Rust
Command-line interface (CLI) for external clients
Native Slint desktop client (Argentum app)
Workspace-scoped provider profiles (LM Studio, OpenAI, Anthropic)
Policy-backed execution profiles (Read Only, Confirm Before Changes)
Bounded session trajectory: tasks, tool calls, approvals, usage
Apache-2.0 open-source license
No subscriptions, telemetry, or cloud lock-in
16 crates across domain, runtime, providers, tools, security, storage, scheduler, platform, UI
JSONL protocol support for external clients
Self-hosted architecture for data sovereignty
Explicit execution policies for agent actions
Developed from scratch in Rust for performance
Supports integration with various LLM providers
Transparent tracking of agent activities
How to use Argentum?
Download & Install: Obtain the Argentum application or CLI for your operating system.
Configure Providers: Set up workspace-scoped provider profiles for services like LM Studio, OpenAI, or Anthropic.
Define Policies: Establish execution policies such as "Read Only" or "Confirm Before Changes" for agent actions.
Create Agents: Develop and integrate your AI agents within the modular Rust framework.
Run Tasks: Execute tasks via the CLI or the native desktop client, observing the bounded session trajectory.
Monitor & Manage: Track agent activities, tool calls, approvals, and resource usage.
Customize & Extend: Utilize the open-source codebase to customize and extend functionalities as needed.
Argentum's Use Cases
- Building Custom AI Agents
- Local AI Agent Development
- Secure Agent Operations
- Open-Source AI Projects
- Privacy-Focused AI Solutions
- Developer Tooling for AI







