CompNav · Open-source projects
Miru
A self-hosted AI companion with a persistent Live2D desktop character, memory, proactive interaction, and multi-device sync.
What it is
Miru presents an AI companion as a desktop character rather than a conventional chat page. With permission, its Live2D character can sense work context, turn relevant summaries into inspectable memory, and use an attention system to decide when to speak. It can run on one device or on a user-owned server shared by multiple devices. The project says it has no central Miru cloud; model requests go to the provider configured by the user.
Core mechanisms and capabilities
The character can use screen summaries and organize past events into memories about people, topics, journals, or commitments; analyzed screenshots are then discarded. An attention system decides whether to initiate conversation based on context. The project supports desktop use and synchronization through a user-owned server.
Editorial take
It merits inclusion because persistent desktop presence, inspectable long-term memory, and proactive conversation form a distinct companion experience rather than merely reskinning a chat window. Screen awareness is sensitive and should be understood and controlled by the user. Several model roles still need API configuration, and clients not listed by the project should not be assumed supported.
Who it suits
Suited to people who want a desktop character to remember daily context and initiate interaction selectively and are comfortable configuring model services, screen-awareness permissions, or self-hosting.
Deployment and technical requirements
The project lists macOS, Windows, and Android clients plus a Linux server mode. Single-device use starts from a release build, while server mode follows the deployment guide and lets devices sign in to sync. Both modes use model APIs selected by the user; there is no official hosted Miru cloud.
Runtime requirements
Users configure compatible model APIs for Vision, Chat, and Memory. Single-device mode runs locally, while cross-device mode requires deployment on a user-owned Linux server. Screen awareness requires permission, and Live2D components or sample characters have third-party terms.
Users configure models for Vision, Chat, and Memory and can mix compatible providers. The README's suggested daily cost is the author's reference, not a fixed project price; actual cost depends on model, usage, and provider.
Limitations and considerations
Screen awareness sends the current screen to the configured vision model for analysis. Although the project says raw screenshots are discarded afterward, users should still consider the provider's data policy. Memory, Chat, and Vision require model configuration, with costs and capabilities set by providers. Live2D asset terms are separate from the app's Apache-2.0 license.
Official sources and further reading
Information checked: Oct 26, 2026.