tersesystems.com

What It Is

Terse Systems is not a conventional developer-tool website in the usual sense. This page is more of the author’s technical blog and personal project log around LLM Agents. Its core project, recipellm, is publicly available on GitHub and can be run with Docker Compose plus API keys for underlying models. The goal is to build a “chef Agent” that can search for recipes, import them into Mealie, add cooking notes, generate shopping lists, and gradually support meal planning and proactive reminders.

Core Capabilities and Tech Stack

The article focuses on the integration between Letta and Mealie. Mealie is described as an open-source recipe manager with a Swagger API, supporting recipe creation from URLs, recipe search, content retrieval by slug, tag and category updates, adding notes, and reading meal plans. Letta handles stateful Agents and function calling. The author finds its ADE useful because it allows Python code to be pasted directly and functions to be debugged interactively. The setup can also connect to Tavily for web search, use ntfy for cooking reminders, and experiment with non-fixed-cycle notifications via Prefect workflows.

Pricing and Deployment

The article does not provide commercial pricing for Terse Systems or recipellm. Actual costs come from third-party models and APIs, such as Gemini, Claude, ChatGPT, Tavily, and others. Deployment details are relatively clear for self-hosting: the project can be run with Docker Compose, while the author’s local setup uses Proxmox and Tailscale, with ntfy connected to an iPhone. It is friendly to developers who value local control and composable toolchains.

Pros and Cons

The main strength is that the content is highly practical, covering real-world issues such as function calling, model provider compatibility, context windows, 429 rate limits, and the limitations of local models. The author explicitly notes that Letta’s OpenAI proxy endpoints are not officially recommended, and that using model providers directly is more reliable. The downside is that this is not mature product documentation: it lacks a complete installation guide, permission model, stability guarantees, and support information. Automated meal planning and proactive conversations are also still exploratory.

Who It’s For and Access from China

It is best suited to developers familiar with Docker, API keys, LLM tool calling, and self-hosted services, either as a reference for Agent integration or as a personal automation template. It is less suitable for general users looking for an out-of-the-box, low-code setup. The article does not mention access from China. However, because the setup depends on GitHub, some overseas model APIs, Tavily, and other services, users in China will need to verify network connectivity, payment options, and API availability themselves. Alternative directions include LangChain, LlamaIndex, Dify, Flowise, n8n, and native tool-calling solutions from model providers.