Mira Oss
Mira Oss: This is the public release of MIRA OS. Discrete memories decay through momentum loss, tools auto-configure when dropped into tools/ folder, and the system prompt composes from modular trinkets. I would like to think I've made an elegant brain-in-box. You load it and send cURL requests - it talks bac
Last verified:
What is Mira Oss?
Mira Oss: MIRA (Memory-Integrated Reasoning Assistant) is an open-source architecture for AI persistence and self-directed context management. Unlike typical AI chat interfaces with a "new chat" button, MIRA maintains one continuous conversation forever, building on every interaction and learning your context over time. It started as a recipe generator that could remember cuisine preferences but evolved into a comprehensive attempt at building a continuous digital entity.
MIRA features a self-managing memory system that automatically extracts memories from conversations. These memories decay over time based on how often they're accessed and how connected they are to other memories—unused information fades while important things persist. The decay formula runs on activity days, not calendar time, so heavy and light users experience equivalent freshness relative to their engagement patterns. Memories can split when too verbose, consolidate when overlapping, and form typed relationships with each other.
For information that shouldn't decay, MIRA uses domaindocs: hierarchical, version-controlled documents for reference material, personal context, and ongoing projects. MIRA controls its own context by expanding and collapsing domaindoc sections as needed, with cost indicators showing what it's trading. Every account has a special domaindoc called personal_context that MIRA writes to directly—a self-model scratchpad for behavioral patterns, preferences, and observations about how interactions actually go.
MIRA is designed for users who want AI continuity and an evolving sense of self. It's ideal for people who tired of repeating themselves to AI assistants, researchers exploring AI persistence and consciousness, developers building persistent AI agents, and anyone wanting a digital entity that truly remembers who they are. The open-source version runs on any Linux or MacOS system with a simple cURL command, with all data staying in your own PostgreSQL instance.
Mira Oss pricing
Pricing model: Freemium
The open-source version is free to self-host with no costs beyond your own infrastructure. You deploy it on any Linux or MacOS system with a simple cURL command for the deploy script, and all your data stays in your own PostgreSQL instance. The hosted version at miraos.org allows you to create an account for access without downloading/installing locally. The hosted version includes a web interface, native MacOS app, and authentication plumbing that the open-source version lacks. No specific subscription prices are publicly listed on the website.
Mira Oss pros
- One continuous conversation forever—no new chat button
- Memory extracts automatically from conversations
- Self-managing memory system with smart decay
- Decay based on activity days, not calendar time
- Memories split when verbose, consolidate when overlapping
- Typed relationships between memories in graph architecture
- Domaindocs for persistent, version-controlled knowledge
- personal_context self-model that MIRA writes to directly
- MIRA documents what works through self-observation
- Fully open-source with identical functionality to hosted version
- All data stays in your own PostgreSQL instance
- Simple cURL command deployment on Linux or MacOS
- Tools auto-configure when dropped into tools/ folder
- Event-driven modules for efficient operation
- Hybrid search for memory retrieval
- Token-efficient tool system with dynamic loading
- Native MacOS app available for hosted version
- Nice web interface for hosted version
- Autonomous context window management
- Built-in CLI tool for interaction
Mira Oss cons
- No way to start fresh—must live with what MIRA becomes
- Requires PostgreSQL installation for self-hosting
- Only has MacOS native app (no Windows/Linux native app)
- Web interface only available on hosted version
- Authentication plumbing only in hosted version
- Complex system—10,000 scope creeps from original recipe generator
- Decay formula may cause important memories to fade if unused
- Self-model writes directly may raise privacy concerns
- Requires Linux or MacOS for deployment (no Windows support)
- No cloud hosting option mentioned for teams/enterprises
Frequently asked questions about Mira Oss
What is MIRA?
MIRA (Memory-Integrated Reasoning Assistant) is an open-source architecture for AI persistence and self-directed context management. It maintains one continuous conversation forever without a new chat button, extracts memories automatically from conversations, and manages its own memory with decay based on activity and connections. It includes domaindocs for persistent knowledge and a personal_context self-model.
How does MIRA's memory decay work?
MIRA's memories decay based on how often they're accessed and how connected they are to other memories. Unused information fades while important things persist. The decay formula runs on activity days, not calendar time—so if you take a two-week vacation, memories don't rot. Heavy and light users experience equivalent freshness relative to their own engagement patterns.
What are domaindocs?
Domaindocs are hierarchical, version-controlled documents in MIRA that hold information that shouldn't decay, such as reference material, personal context, and ongoing projects. MIRA controls its own context by expanding and collapsing domaindoc sections as needed, with cost indicators showing large sections so MIRA knows what it's trading when expanding something.
What is the personal_context domaindoc?
The personal_context is a special domaindoc that every MIRA account has. MIRA writes to it directly as a scratchpad for behavioral patterns, preferences, and observations about how interactions actually go. MIRA documents what works, what doesn't, and who it is through watching how it operates and annotating discoveries, not through external configuration.
How do I deploy MIRA locally?
MIRA is deployable on any Linux or MacOS system with a simple cURL command for the deploy script. The open-source version is effectively identical to the hosted version except for web interface and authentication plumbing. All your data stays in your own PostgreSQL instance when self-hosting.
What's the difference between open-source and hosted versions?
The open-source version is effectively identical to the hosted version in core functionality. The differences are that the hosted version includes the web interface and authentication plumbing. The open-source version stores all data in your own PostgreSQL instance, while the hosted version runs on miraos.org with account-based access.
Can I use MIRA without installing it locally?
Yes, you can create an account on miraos.org to access the hosted web interface without downloading and installing on your local computer. The hosted version also includes a native MacOS app for those who prefer not to use the web interface.
How does MIRA extract memories?
MIRA extracts memories passively and automatically from conversations. These discrete memories are then paired with larger text segments that the model can autonomously edit. The system uses graph-based architecture with hybrid search for memory retrieval and self-maintenance.
Why is MIRA open source?
The creator released MIRA as open source because they believe what they've built has potential to someday be something more than the sum of its parts, and no one person should own that. The open-source release allows anyone to run their own instance with full control over their data.
What tools and interface options does MIRA offer?
To interact with MIRA, you can use /v0/api/chat endpoint or the CLI tool. The hosted version offers a nice web interface and a native MacOS app. Tools auto-configure when dropped into the tools/ folder with dynamic loading and token efficiency. You can check system health by querying /v0/api/health.