Cloud trial and private deployment
Validate first, choose the deployment shape second. The cloud version is for trials and evaluation; production can be self-hosted or moved onto your own network.
Cloud trial
Sign up for the cloud version and start; there is no server to prepare first. It suits evaluating what the platform can do and checking whether a particular business case holds up.
Self-hosted
A single process with SQLite and no external dependencies, on Python 3.11+ or Node.js 18+. LLM_API_KEY is the only required setting.
For production, Docker Compose deploys in one command and configures PostgreSQL for you. As the deployment grows you can migrate from SQLite to PostgreSQL and scale out to multiple workers.
On-premises private deployment (planned)
For government, finance, and other settings with strict data-residency requirements: all dependencies install offline, the system runs disconnected from the internet, and it targets domestic Chinese hardware and OS stacks.
Bring your own model
Any provider exposing an OpenAI-compatible /v1/chat/completions endpoint works. Change LLM_BASE_URL and LLM_MODEL to switch; the business flows stay as they are.
Models can also be assigned by role (general, fast, reasoning), with the agent switching between steps as needed.