LEANN
[MLsys2026 Best Paper]: https://arxiv.org/abs/2506.08276. RAG on Everything with LEANN. Enjoy 97% storage savings while running a fast, accurate, and 100% private RAG application on your personal device.
Runs with mockssource: GitHubhomepagePythonMITcommit f84dec41db5b
Python, MIT licensed. The project labels itself: ai, faiss, gpt oss, langchain, llama index, llm, localstorage and offline first.
LEANN runs, with stand-ins for the services it depends on. An Argusic agent installed it in 85 minutes and hit 8 errors on a clean machine with no GPU, and the whole session was recorded.
LEANN Python packages installed, HNSW backend with custom-built Faiss SWIG module works, CI tests pass, README HNSW examples pass with mock embeddings, DiskANN backend blocked by missing libaio system dependency.
What the agent ended up with on a clean machine, in its own words. How this is measured
Measured by Argusic on a fresh machine every time. Every number links to its evidence. Argusic Score 89.5 of 100 (the mean of this project's run scores).
At a glance
Subject data from GitHub, linked at the top of this page, refreshed . Test data by Argusic (CC BY 4.0); every number links to a run page with the full log, the recording, and their sha256 hashes.
What was measured
From the run that decided this verdict. Every line is a recorded value, not an estimate.
What went wrong, in the agent's own words
The agent's own words, worst first, never rewritten into advice.
- Faiss Python wrapper files had circular import structure when nested under leann_backend_hnsw10 minutes
- leann-backend-diskann requires libaio and Boost system packages (no root access)10 minutes
- CMake FindBLAS could not find BLAS (only libblas3 runtime, no .so symlink without version)5 minutes
- CMake could not find SWIG during Faiss build5 minutes
- Faiss SWIG module compiled with numpy 1.x but env had numpy 2.x, causing SystemError at import5 minutes
- Git submodules not initialized3 minutes
- Install time
- 85 minutes
- Cold machine to finish
- 87 minutes
- Errors hit and fixed
- 8 hit, 7 fixed with no human help
- How the result was proved
- pytest tests/test_ci_minimal.py -v: 4 passed, pytest tests/test_readme_examples.py -v -k hnsw: 2 passed
- Model tokens used
- 1,755,633
- Exact commit tested
- f84dec41db5b
- Container image
- sha256:cdd920bce783
What was not measured
The other half, so you can see exactly how far this evidence goes. A gap is printed as a gap and never filled in.
- Peak GPU memory
- this ran on a CPU machine, so there was no GPU to sample
- Behaviour over a long run
- each attempt uses the smallest documented settings and a fixed time budget, so this measures whether it runs, not how it holds up under load
- Behaviour against real services
- this run was verified against stand-in services only
The end of the recording
The last lines of the terminal session that decided this verdict, exactly as they were recorded.
2026-09-26T20:42:47.957373Z ERROR codex_core::session: failed to record rollout items: thread 01a0d… 2026-09-26T20:42:47.957387Z ERROR codex_core::session: failed to record rollout items: thread 01a0d… 2026-09-26T20:42:47.957399Z ERROR codex_core::session: failed to record rollout items: thread 01a0d… 2026-09-26T20:42:47.957412Z ERROR codex_core::session: failed to record rollout items: thread 01a0d… 2026-09-26T20:42:47.957425Z ERROR codex_core::session: failed to record rollout items: thread 01a0d… 2026-09-26T20:42:47.957436Z ERROR codex_core::session: failed to record rollout items: thread 01a0d… 2026-09-26T20:42:47.957451Z ERROR codex_core::session: failed to record rollout items: thread 01a0d… 2026-09-26T20:42:47.957466Z ERROR codex_core::session: failed to record rollout items: thread 01a0d… 2026-09-26T20:42:47.957480Z ERROR codex_core::session: failed to record rollout items: thread 01a0d… 2026-09-26T20:42:47.957499Z ERROR codex_core::session: failed to record rollout items: thread 01a0d… tokens used 1,755,633
Replay the whole session, every command from a clean machine to this point.
How it was tested
One agent, one container, no human help. It installs the project the way its own documentation says to, runs it, and fixes what breaks. Everything below is recorded as it happened: the terminal session, the log and the exact commit. The full method.
Strengths and limits
Measured facts, not opinions. How this is written.
What went well
- Licensed MIT, as reported by its host.
What did not, or is not known
- Only reached a running state against stand-in services, so nothing here proves it works against the real ones.
- Took 85 minutes to install, slower than the median of the 46 comparable projects Argusic has measured.
- Hit 8 errors during setup, 1 of which were not solved.
- Measured once so far. A single attempt shows it can run, not that it runs reliably.
Of the 47 ai projects Argusic has installed and timed, LEANN was the 47th fastest to reach a running state, and 37 of 47 reached one at all.
Also tested, in the same area
Every one of these was installed and run by Argusic on a clean machine. Nothing appears here that was not tested.
Run history
| tested by | attempt | outcome | score | cost USD | when |
|---|---|---|---|---|---|
| Argusic Runner | 1/3 | Runs with mocks | 89.50 | 0.54 |
Topics (from GitHub)
aifaissgpt-osslangchainllama-indexllmlocalstorageoffline-firstollamaprivacypythonragretrieval-augmented-generationvector-databasevector-searchvectors
Embed the badge
Markdown for the project README. It links back here; terms on the terms page.
[](https://argusic.com/subject/leann)Questions
- Does LEANN run?
- LEANN runs, with mocks standing in for external services. Argusic installed and launched it on a clean machine in 85 minutes, hitting 8 errors on the way, and recorded the session.
- How did Argusic test LEANN?
- On a fresh, disposable machine, with every command recorded and the repository pinned at commit f84dec41db5b. One attempt is recorded, and the full method is on the methodology page.
- What did it cost to test LEANN?
- The run that produced this verdict cost $0.54: real compute and model cost, not a list price per million tokens. It is recorded on that run's page.
- How long does LEANN take to install?
- 85 minutes in the run that decided this verdict, from a clean machine to a running state. That is one measured attempt, not an average.
- Does LEANN need a GPU?
- Not to start. Argusic reached a running state on a machine with no GPU.
- What goes wrong when installing LEANN?
- 8 things broke in the recorded run, and 7 were fixed without human help. Each one, and the time it cost, is listed on this page.
- How does LEANN compare with the alternatives?
- Of the 47 ai projects Argusic has installed and timed, LEANN was the 47th fastest to reach a running state, and 37 of 47 reached one at all.
- Where is the evidence for LEANN?
- The recorded run is on this page, each linking to its full log and terminal recording, stored with a sha256 fingerprint so it cannot be quietly altered.