This page is a dated snapshot, taken on 2026-10-21 and written against ragleap-core v0.11.0. It lists what has shipped, what is experimental, and what is planned but not built. Plans have no dates and are not promises. For the exact contents of each release, read the release notes. Unreleased work on the main branch is not listed here until a release contains it.
Released in v0.11.0 (2026-10-20)
- Documents tab. List, upload and delete documents. Deleting also removes the document's entries from the knowledge graph.
- 28 file types and web pages by address can be added from the dashboard.
- Proactive and Connectors tabs in the AI Office.
- Offline language detection.
LANGUAGE_DETECTION_MODEL=liteworks without internet. - Safer container. The app runs as a non-root user with pinned dependencies.
- One scheduler. Only one process runs the background jobs.
- One more provider option. An OpenAI-compatible hosted service, chosen with
LLM_PROVIDER.
Earlier releases (login throttling, the Setup tab, free ports and key rotation, the AI Office dashboard, agents and the installer) are described in the blog.
Experimental, off by default: computer control
The first slices shipped in v0.11.0. Folders can be shared with AI employees only by a command run on the machine (ragleap files). A sandbox runner reads them read-only, and every access request needs the owner's approval. It has not been independently reviewed. Please do not switch COMPUTER_CONTROL on for real data. Reviewers are welcome: issue #800.
Planned, not built (no dates)
None of these exists today. They are directions, listed so you can see where the project wants to go and tell us what matters to you. Anything that acts for you would go through the same approval settings as every other action.
- Named users and roles, with a person recorded in every audit entry.
- An offline install bundle.
- Metrics and tracing.
- A permission system for local files and folders (the first slices are the experimental computer control above).
- Parallel research with sources.
- Generating Word, Excel and PowerPoint files.
- Spell check.
- Read-only analysis of business data.
- Drafts of ads and social posts, published through official APIs only after approval.
- Browser control in an isolated profile.
- Voice and an AI presenter, starting with narrated slides.
Known limits today
- No named users or roles yet, and no independent security review yet.
- The compose file still ships a public default password for the graph database (Neo4j listens on localhost only). A random password per install is planned.
- Tested by the maintainer on one Windows laptop and on Linux servers. macOS, Linux desktops and the Ollama option on Windows are not verified yet (issues #779 to #781).
Tell us what matters
Open an issue or start a discussion. The issues page shows what has been resolved and by whom.