Install Provon
You can use Provon as a hosted service or run it yourself. This page covers all three paths.
| Path | Best for | Model fine-tuning |
|---|---|---|
| Hosted Provon | Fastest start, no infrastructure | No |
| Self-hosted installer | Private data, local models | Yes |
| Source development | Contributing or customizing Provon | Yes |
Hosted Provon#
The fastest way to start:
- Open
https://app.provon.dev. - Sign in and complete the Workbench onboarding.
- Copy the project API key shown during onboarding.
No installation is required. The hosted service does not support local model serving or fine-tuning; for those capabilities, use a self-hosted runtime.
Self-hosted installer#
The official installers download a release archive, set up a local Node server, Python inference
environment, and data directories, and generate a ready-to-use env file.
Requirements#
- Node.js 22 or newer (the installer can download a private copy if needed)
- macOS, Linux, or Windows
- An empty directory for the Provon prefix and data
macOS / Linux#
curl -fsSL https://provon.dev/install.sh | shWindows#
irm https://provon.dev/install.ps1 | iexWhat the installer does#
- Resolves the latest Provon release (or use
--version/-Version). - Downloads the platform-specific archive from GitHub Releases.
- Verifies the SHA256 checksum.
- Extracts the CLI, Node server, Python inference wheel, and fine-tuning wheel.
- Ensures Node.js 22+ is available.
- Creates a Python virtual environment and installs the wheels with the selected engine extras.
- Generates
env(andenv.ps1on Windows) with random secrets and local paths. - Optionally adds the
bindirectory to PATH.
Common installer options#
Shell (install.sh) |
PowerShell (install.ps1) |
Default |
|---|---|---|
--prefix PATH |
-Prefix PATH |
~/.provon / %USERPROFILE%\.provon |
--engine EXTRA |
-Engine EXTRA |
llama_cpp |
--fine-tuning-engine EXTRA |
-FineTuningEngine EXTRA |
none |
--data-dir PATH |
-DataDir PATH |
<prefix>/data |
Engine extras:
- Inference:
llama_cpp,vllm,sglang,all,none - Fine-tuning:
transformers,gguf,compressed,all,none
Start the self-hosted server#
After installation, follow the installer output to source the environment file and run:
source /path/to/provon/env
provon-serverThen open the Workbench URL printed in the logs (defaults to http://127.0.0.1:3000).
See Self-hosting overview for architecture, configuration, security, and operations.
Source development#
Use this path if you are contributing to Provon or need to customize the source.
Requirements#
- Node.js 22 or newer
- pnpm 10.33.0 or newer
- Python 3.11+ if you plan to run local model services
Clone and install#
git clone https://github.com/provon/provon
cd provon
pnpm installStart the local development server#
pnpm devThis starts:
- the API server on
http://127.00.1:3001 - the Workbench dev server on
http://127.0.0.1:3000 - a TypeScript watch compiler for
services/node-server
Local data is stored under .provon/ in the repository root:
- metadata in libSQL/SQLite
- telemetry in DuckDB
- blobs on the local filesystem
Use a custom Workbench port when needed:
pnpm dev -- --port 3100Seed a local product tour#
To inspect the Workbench without generating your own workload, seed an empty local database before starting the server:
pnpm seed:local
pnpm devThe seed command prints the test account, project API key, and inserted record counts. It refuses to write into non-empty metadata, metering, or telemetry databases.
Install the CLI only#
The self-hosted installer already includes the CLI. If you only need the CLI against an existing Provon deployment:
# macOS / Linux
curl -fsSL https://provon.dev/install.sh | sh
# or from the repository
pnpm --filter @provon/cli build
node cli/dist/bin.js --helpAuthenticate with a project API key or with user delegation:
# Project API key
export PROVON_API_KEY="your_project_api_key"
export PROVON_API_URL="https://api.provon.dev/v1"
# Or log in through the browser
provon auth loginSee CLI authentication and CLI overview.
Verify the installation#
Run:
provon --versionFor a self-hosted or source install, check that the Workbench loads and that you can create a project. For any install path, the next step is the Quickstart.