Experiments with nachalnik
Show HN: What an agent does when anyone can read and rewrite its context
Last verified:
Visit Experiments with nachalnik
What is Experiments with nachalnik?
Nachalnik is an open-source, Rust-based agent runtime designed to prioritize transparency by treating an AI agent's context, tools, and permissions as explicit, mutable state. Aimed at developers and researchers, it includes a terminal user interface called Kamchatka that lets users actively monitor, edit, and sandbox the agent's execution.
Experiments with nachalnik pricing
Pricing model: Freemium
Free (Open-source)
Experiments with nachalnik pros
- Provides full transparency by allowing users to inspect and manually rewrite an agent's working memory
- Includes Kamchatka, a terminal user interface for managing context, pinning items, and controlling requests
- Supports local and hosted models via OpenAI, Anthropic, and Gemini APIs
- Features an MCP (Model Context Protocol) bridge and a securely sandboxed shell using Landlock and seccomp
Experiments with nachalnik cons
- Lacks a graphical user interface, operating strictly as a command-line and terminal tool
- Highly technical setup geared toward developers, requiring command-line familiarity
- As an abstract framework, it has a steeper learning curve compared to standard black-box AI assistants
Frequently asked questions about Experiments with nachalnik
What AI models are supported?
Nachalnik supports models that use OpenAI's chat completions, Google's Gemini API, Anthropic's Messages API, as well as local models via Ollama, vLLM, or LM Studio.
How does the tool handle secure shell commands?
It uses a shell sandboxed by the kernel (Landlock) and a seccomp filter to intercept network requests, requiring user approval before proceeding.