Local-first Studio

Run SILT Studio on your machine.

The public SILT site is the discovery and launch surface. The actual Studio runs on your own computer at 127.0.0.1:8377, so your models, skill packets, benchmark data and audit state do not need to pass through Vercel.

Open Local Studio View GitHub

If the local Studio is already running, the first button opens it immediately. Otherwise install/start it below.

Privacy boundary: 127.0.0.1 always means the visitor's own machine. SILT Studio binds to loopback by design; the public website does not expose a user's local model runtime to the internet.

Install SILT

Clone the open-source repository and install the Studio extra.

Run one command

silt-studio checks whether Studio is already running, starts it if needed, binds only to loopback, and opens the browser.

Work locally

Configure teacher and learner systems, run transfers, inspect gates and audit evidence without moving the local execution layer to Vercel.

Install and launch

Windows · PowerShell

git clone https://github.com/inbharatai/SILT.git cd SILT py -m pip install -e ".[studio]" silt-studio

macOS / Linux · Terminal

git clone https://github.com/inbharatai/SILT.git cd SILT python3 -m pip install -e ".[studio]" silt-studio

Requires Python 3.9 or newer. Real-model connector dependencies and deep-apply dependencies remain optional and can be installed when those features are needed.

Choose a local workflow

Legacy Studio: Transfer an instruction packet, Train approved packets into a removable adapter, or Compress for a quantized-state loss report. The optional workbench adds composition, smaller specialists and evidence inspection; CLI/library-only paths remain in the capability catalog.

Experimental workbench — disabled by default. This public page does not inspect or enable your local server. Stop the existing Studio before restarting with the exact flag SILT_ENABLE_EXPERIMENTAL=1. Unset, 0 and true stay off; setting the flag after an off-start requires restart. The launcher reuses a running server, so changing only its environment is insufficient.

macOS / Linux · from the repository

SILT_ENABLE_EXPERIMENTAL=1 PYTHONPATH=src python -m uvicorn asea.studio.server:app --host 127.0.0.1 --port 8377

Windows · PowerShell

$env:SILT_ENABLE_EXPERIMENTAL="1"; $env:PYTHONPATH="src"; python -m uvicorn asea.studio.server:app --host 127.0.0.1 --port 8377

Open Local Studio above, then use its enabled-workbench link. No credentials or session tokens are sent to this public page. Read workbench prerequisites and recipes · Specialist guide.

App available ≠ model ready. The Studio extra starts the app only. Default HF teacher/learner and embedding similarity require connector dependencies ([connectors]) and assets; Train/Compress additionally need [deep]. Before executing, confirm model assets, endpoint location, device fit and budget. A loopback Ollama endpoint can use cloud-backed tags (for example glm-5.2:cloud); prompts may leave your machine and incur cost. This setup page performs no downloads or provider probes. Dependencies can require newer Python than the package minimum.

What the launcher does

✓ checks http://127.0.0.1:8377/api/health ✓ reuses an already-running SILT Studio ✓ refuses the port if another service owns it ✓ starts SILT only on 127.0.0.1 ✓ opens http://127.0.0.1:8377/ automatically ✓ never adds a public 0.0.0.0 bind option

Already running?

Your Studio address is:

http://127.0.0.1:8377/

Launch SILT Studio

Why the execution layer stays local

SILT Studio works with local model runtimes, local files, benchmark suites, skill packets, audit state and potentially large model weights. Hosting the documentation and launcher is lightweight; sending every user's private AI workload through a public cloud service would change SILT's security and privacy model. The local-first architecture keeps compute and data where the user chooses to run them.