Alma, provided by Olivares AI, is a self-hosted, vendor-neutral AI runtime and governance platform, adopting an open-core model, licensed under AGPL-3.0, and hosted on GitHub. Its core goal is to let users run AI tools on their own infrastructure while governing the content they connect to.
Purpose and Positioning The platform can run coding sessions on your own infrastructure, suitable for a variety of environments from home servers to regulated enterprises. It is officially stated not to be tied to any single AI vendor, identity provider, or cloud service, and the Community edition can run without mandatory calls home; installation, updates, and external providers have their own connectivity requirements.
Main Capabilities The platform organizes its workflow around four steps:
- Deploy: Install Olivares AI onto your own infrastructure as a binary or container, using SQLite or PostgreSQL for storage.
- Discover: Connect supported observation sources, and the Inventory reflects the agents, sessions, MCP servers, and models reported by these sources.
- Map: Based on connected sources, build a map of allowed access and actually observed access. Read and write details depend on the evidence provided by these sources.
- Govern: Configure access policies, inspect recorded events, and export evidence. Policies take effect when sessions or tools run through Olivares AI.
The following capabilities are also officially mentioned: real-time operations and sessions, allowing inspection of observed session operations, models, tokens, costs, and event history, and including governed work items and handoffs; the reported goals depend on the reporting sources, and it is explicitly stated that calculated task progress and complete operational workflows have not yet been verified. One-click estate stop and dual-control recovery are provided. For compliance evidence, control status and sealed evidence are provided, and it is officially emphasized that this is not certification. In addition, model spend can be viewed, attributed, and governed.
How to Use The officially given path is: install Olivares AI → connect supported sources → inspect the activity they report. The platform works with existing tools, and visibility and enforcementeffects depend on the specific integrations and the controls configured by the user. A product tour is officially provided (including real console screenshots) as well as a video explanation introducing how the parts of the project fit together, with the video on YouTube.
Versions and Licensing Version 1.0 is officially being prepared, the first stable release of Community and Business/Enterprise. The Community edition under AGPL-3.0 is a complete product for individuals or teams, can be self-hosted, and its source code can be viewed; the paid versions add capabilities, commercial licensing, and support. It is also officially mentioned that sponsorships are used to sustain the development, integration, and updates of Olivares AI; it is not a support contract and does not purchase priority.
Limitations and Considerations In the official description, read and write access and least-privilege drift are described as differentiating features. Visibility and enforcement depend on integrations and control configuration. Calculated task progress and complete operational workflows have not yet been verified. Compliance control mapping and evidence are not certification. The sample data in the demo screenshots does not represent any customer's compliance status. Details such as which observation sources are specifically supported, pricing, platform and system requirements, etc., are not yet provided in the materials.
