RagLeapEverything on this page is Planned unless a label says otherwise. Nothing here is a feature you can use today, and nothing has a date, a device count or a performance number, because none of that has been measured.
Today
RagLeap Core is free, MIT-licensed, self-hosted software with 46 ready-made AI employee roles. By default they take no action on their own, and in approval mode every real action waits for the owner’s yes. It is single-tenant today. What exists is described on the roadmap, with its limits.
Long-term direction (Planned)
A Universal Intelligence & Orchestration Platform: governed AI that connects organizations, people and AI employees, with explicit authority, privacy, accountability and decisions backed by evidence. For physical systems the rule is simulators only, for now.
- Organization types and branches, kept separate (Planned). Different kinds of organizations, with branches and departments that keep their data separate by default, and coordination between them only where it is authorized.
- Shared human and AI projects (Planned). People and AI employees working on the same project, with handoffs, reviews and a timeline that shows who or what did each task.
- An assistant for a person, with their consent (Planned). An AI assistant that helps one person with their own work and knowledge, only with that person’s consent, and never acting as that person.
- Operational intelligence, with evidence (Planned). Patterns across applications and time, shown with the supporting evidence, what is missing and how uncertain it is. A suspicion is never presented as a fact.
- An executive view (Planned). One place where authorized people see progress, open issues and decisions, with drill-down to the evidence they are allowed to see.
- A connector framework (Planned). Third-party applications connect through connectors that declare what they can read, which commands they may run, their risk level and which approvals they need.
- Outcome verification (Planned). After a recommendation or an approved action, check whether the intended outcome happened, and record failures and human corrections.
- Acceptance tests before claims (Planned). A capability is declared supported only after its acceptance tests pass.
Physical systems
Released, pre-alpha, simulators only: ragleap-physical lets an AI propose a change while a named person approves the exact change, tested in simulation first. Nothing in it controls real hardware. Life-safety systems, medical devices, vehicles on public roads and weapons are out of scope, always.
The status words we use
- Planned: no code yet.
- In design: a public design issue exists.
- In development: an open pull request.
- Merged: in the main branch, not released yet.
- Released: a tagged release or a published package.
- Community-tested (one environment): a community member ran it on one setup and published the evidence; not a compatibility guarantee.
Public design discussions are on GitHub. For what has shipped and what is known to be missing, read the roadmap.