Tools for a reproducible AI research loop.
Website · Getting started · Documentation · Learn the field · Brand assets
darsay helps engineers build and operate an AI research loop from open weights: understand the model, run controlled experiments, evaluate changes, and build on the results. The project's guides, tools, services, and hardware reference designs support that work. The north star is to improve the models and research process used in each successive cycle, working toward recursive self-improvement as a research goal.
This repository contains the darsay CLI, which preserves the inputs to that loop. It catalogs models, datasets, and source repositories as pinned, checksummed bundles in a vault you control. Files keep their published formats, with licenses, provenance, and verification evidence recorded beside them. Supported models run offline in isolated environments; collections and a shared browser interface help people and agents coordinate acquisition.
The CLI direction is to connect existing training, inference, and evaluation tools through inspectable, removable project metadata. General loop orchestration is proposed. Today, use the CLI for artifacts and supported runtimes, and your experiment's native tools to train, evaluate, and retain the research record.
Install
Python 3.10+ on macOS or Linux:
pipx install darsay
# or
uv tool install darsay
Try the CLI without a permanent install with uvx darsay --help.
Installation and upgrades covers
Homebrew and other installation paths.
Quick start
darsay estimate Qwen/Qwen3-0.6B
darsay archive Qwen/Qwen3-0.6B
darsay run qwen--qwen3-0.6b "Say hello"
The default vault is ~/darsay; use --vault /path/to/vault to choose another.
Interrupted archives resume with the same command. The first run prepares its
runtime environment; inference then uses the archived files offline.
The guided first bundle explains each step.
Browse an existing vault:
darsay --vault ~/darsay serve --open
Open the private launch URL and use All archives or a named collection. The vault UI guide covers local operation, read-only browsing, collection exports, and agents.
Documentation
The full guides and reference live at darsay.io/docs.
| You want to… | Read |
|---|---|
| Understand the research loop and the CLI's role | Project north star · CLI direction |
| Connect a preserved model to training and evaluation | A model improvement loop |
| Understand vaults, bundles, and collections | Concepts |
| Find a command or workflow | Examples |
| Curate and coordinate a collection | Catalogs · Boards and agents |
| Resume, move, or back up archives | Incremental transfer · FAQ |
| Understand and run a model | Learn the field · Hydration |
| Recover an archive independently of darsay | Manifest · MVB format |
The CLI documentation's source remains in docs/ and
examples/, available offline with this repository.
For development, see Contributing; for changes, see
the changelog.
License
Apache License 2.0. Archived works retain their own upstream licenses, recorded separately in each bundle's manifest.
1 maintainer
Metadata
1 maintainer
Release files for darsay 0.17.7
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | |
|---|---|---|---|
| darsay-0.17.7.tar.gz | 3.3 MB | Details |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Reset |
|---|---|---|---|---|
| darsay-0.17.7-py3-none-any.whl | Python 3 | none | any | Details |
Total release size: 4.4 MB
Release files / darsay-0.17.7.tar.gz
| Download URL | darsay-0.17.7.tar.gz |
|---|---|
| Size | 3.3 MB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
ad64e55a4dcd6abd2ed68392bc46de342453e8d28604814748a3812f6dceb355
|
|
BLAKE2b-256 checksum How to use checksums |
fa757f562f68301339f161d91213caa3774dab39e846024ed868a34316fdd1a5
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Oct 10, 2026.
Transparency logRelease files / darsay-0.17.7-py3-none-any.whl
| Download URL | darsay-0.17.7-py3-none-any.whl |
|---|---|
| Size | 1.1 MB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
2604ed7dd53a62e21ce5c9babd76721ac36fb508c41d02e4e900a15564c19e0d
|
|
BLAKE2b-256 checksum How to use checksums |
6104fd9814f7d3a1ec4300e7d0b619e25a0a9b2e087b2f89ca4131904f14d6a1
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Oct 10, 2026.
Transparency log
