RagLeapShort answer: one community tester, @Anurag07-crypto, ran the RagLeap Core Windows installer with the Ollama option on one Windows 11 Pro machine. The services started, the dashboard opened, the chat and embedding connection tests passed, one PDF was ingested as 5 chunks, and the chat API answered with the source named. The maintainer reviewed the evidence on 2026-11-05 (issue #781). It is one machine, one PDF and one question, so it is a community test of one environment and not a compatibility guarantee.
What was tested?
RagLeap Core installed on Windows 11 with Docker Desktop, using local models through Ollama instead of a hosted provider. Then one PDF was ingested and one question was answered through the chat API. The environment, as shown in the screenshots:
- Windows 11 Pro 25H2, build 26200.9711
- Docker 29.9.2 with Compose v5.6.1, Git for Windows 2.54.0, Ollama 0.40.14
- Chat model qwen2.5:3b, embedding model nomic-embed-text (768 dimensions)
What does the evidence show?
- The services started and the AI Office dashboard opened at localhost:8000/office.
- The chat connection test and the embedding connection test passed (reported in the tester’s results file).
- One PDF (pdh.pdf) was ingested and produced 5 chunks.
- The chat API returned HTTP 200 with an answer, the source pdh.pdf, 5 chunks used and 4 sent to the model, the provider ollama, and 2,613 tokens in total (2,493 for the prompt and 120 for the answer). This is shown in a screenshot of the response.
- The first response was slow. The tester thinks the model was loading, but response times were not measured.
- The dashboard has no chat screen today, so the tester used the chat API for the question.
Where is the evidence?
The summary, the results file and the screenshots are in the community-tests repository. Three silent screen recordings (the installation, the dashboard overview and the chat request) are attached to its release. The installation recording is edited or sped up: it runs about six minutes, while the build step on screen reports about 15 minutes, so do not read its length as an install time. The summary was prepared with AI assistance from the tester’s files, and the original files are linked next to it.
What does it not show?
- Other Windows versions, other hardware or other Docker and WSL setups.
- Install time, model download time or response times.
- Answer accuracy beyond this one PDF and one question, or retrieval quality across documents.
- Repeated runs, large PDFs, several users at once or long sessions.
- Autonomous actions or approvals: autonomy was off, so only answering was tested.
- Security, or an offline install without internet.
What changed on this site?
The installer line on the roadmap and the homepage now says the Ollama option on Windows is community-tested (one environment) and links the evidence. macOS and Linux desktops are still not verified (issues #779 and #780 are listed on the roadmap).
How can I add a test?
If you try the installer on another setup, say so on the issue: write down the operating system and the versions, what you ran, what worked and what did not. Failures are as useful as passes. Thank you to @Anurag07-crypto, who also tested ragleap-physical on Windows (issue #1114).