mycelia.garden

What It Is

Mycelia is a desktop knowledge-base tool for users who “talk to AI every day,” with support for macOS, Windows, and Linux. Its core idea is not to build yet another closed AI memory system, but to automatically turn conversations into a locally controlled Markdown vault: daily logs, structured notes, tags, [[wikilinks]], and a relationship graph. In later conversations, it can retrieve these notes first before generating a response.

Core Capabilities

On the AI side, the free version uses local Ollama models, making it suitable for users who care about privacy and low cost. Pro can connect to Claude, GPT-4, Gemini, 100+ models via OpenRouter, as well as any OpenAI-compatible API, including self-hosted endpoints. Typical use cases include product pricing discussions, feature-priority brainstorming, project decision records, personal work journals, and long-term knowledge retrieval. It is compatible with Obsidian: the vault is essentially a set of Markdown files in a normal folder, which can continue to be managed with Obsidian, VS Code, git, or any text editor.

Pricing and Privacy

Pricing is straightforward: the Free plan is €0, free forever, requires no account, and includes local AI, the full vault, graph view, search, note editing, and daily journals. Early Bird Pro is a one-time purchase of €24, with the page listing the official launch price as €42; it unlocks cloud models, semantic search, and future updates. However, Pro requires users to bring their own API keys, and model usage fees are charged directly by the third-party providers. On privacy, the free version runs locally and data does not leave the device; with Pro, API requests are sent directly from the user’s machine to model providers rather than through Mycelia servers, and API keys are stored locally.

Pros, Cons, and Who It’s For

Its strengths are that it is local-first, uses open formats, avoids lock-in, is Obsidian-compatible, has no subscription, and can automatically turn fragmented AI conversations into searchable knowledge. The downsides are that the product is still in alpha, so there may be bugs and feature changes; Chinese-language support is not specified; support mainly appears to be via X; and while cloud models are powerful, they depend on external APIs and network conditions. It is best suited to developers, indie founders, product managers, research-oriented writers, and heavy Obsidian users. It is less suitable for teams that want a fully hosted, ready-to-use solution with no API configuration required.

Access from China

The page does not provide details on access, payment, or model connectivity from mainland China, so the access status can only be marked as unknown. Stripe payments may be affected by domestic bank card conditions, and services such as Claude, OpenRouter, and OpenAI may also be limited by network and account restrictions in China. If access is restricted, alternatives include using Obsidian/Logseq with local Ollama, or replacing it with China-accessible LLMs combined with a local Markdown workflow.