Deployment
Local machines, Docker, Railway, VPS, and remote execution backends.
Synapse is not tied to your laptop. Start it on a VPS or a GPU cluster and talk to it from Telegram while it works in the background. This page covers where the agent can live and how the pieces connect.
Local
npx synapse-ai-agent
# or the direct installer:
# curl -fsSL https://raw.githubusercontent.com/johsua092-ui/synapse-ai-agent/main/scripts/install.sh | bashNative installs live under %LOCALAPPDATA%\synapse on Windows and ~/.synapse elsewhere.
Docker
SYNAPSE_UID=$(id -u) SYNAPSE_GID=$(id -g) docker compose up -dSupervised setup with s6-overlay entrypoint. Windows variant: docker-compose.windows.yml.
Railway
- One-click template: railway.com/template/synapse-agent (auto-detects
railway.toml+ Dockerfile) - Mount a volume at
/opt/datafor persistent agent state SYNAPSE_DASHBOARD=1enables the dashboard — public binds require Basic Auth or OAuth/OIDC and fail closed without auth- Health check endpoint:
/api/health
VPS
Any modest VPS works — the README calls out a $5 droplet as viable. Install via the script, run the gateway, and reach your agent from Telegram while it works server-side.
Remote execution backends
Seven terminal backends decide where tools actually execute: local, Docker, SSH, Singularity, Modal, Daytona, and Vercel Sandbox. Daytona and Modal offer serverless persistence — the environment hibernates when idle and wakes on demand.
| Backend | Best for |
|---|---|
| local | Everyday use on your own machine |
| Docker | Reproducible, isolated environments |
| SSH | Running tools on a remote box you already have |
| Singularity | HPC clusters |
| Modal / Daytona | Serverless scale-to-zero persistence |
| Vercel Sandbox | Ephemeral preview-style environments |