Synapass Desktop — Portable and Setup builds for Windows

Synapass Desktop v1.0.0 (Windows x64)

The first stable release of Synapass Desktop — the full Synapass inference gateway and control plane, running locally on Windows in its own window. No Docker, no system database, no admin rights required.

Synapass sits between your apps and your model providers behind a single OpenAI-compatible endpoint, and decides per request which provider serves it, what happens when a provider fails, and what it cost.

Assets

File	What it is	Size
SynapassSetup.exe	Setup build — installer-based version	~238 MB
Synapass.exe	Portable build — app only, no installer code	~18 MB

Builds

Portable — runs without installation

• Launch Synapass.exe directly; there is no installer and no setup step.
• Nothing is registered with Windows: no shortcuts, no services, no Add/Remove Programs entry.
• The runtime and all data live beside the executable, so you can keep it on a USB drive, a network share, or any folder you choose.
• Removing it is just deleting that folder.

Setup — installer-based

• Run SynapassSetup.exe and follow the seven-step wizard: Welcome → Install location → Options → Admin account → Review → Install → Finish.
• Choose the install folder and a separate data folder (default: your user profile — another drive works too).
• Options: desktop shortcut, Start Menu entry, start at sign-in, and launch when setup finishes.
• Creates the shortcuts and the Add/Remove Programs entry, then exits. The installer never becomes the app.
• Unattended mode for scripts and CI: SynapassSetup.exe --silent --config install.json.

Both builds run the identical application — pick whichever delivery you prefer.

What's inside

Everything needed to run is bundled:

• Gateway (Go) — OpenAI-compatible /v1/chat/completions (streaming and buffered) and /v1/models, with multi-provider routing, fallback chains, policy enforcement, cost ceilings, budgets and request caching.
• Dashboard (Next.js) — the full operator console: overview, requests with per-request explain, errors, providers, models, policies, endpoints, tenants, API keys, budgets, scores, cache, tools, analytics and tunnels.
• Embedded PostgreSQL — started, migrated and supervised by the app; migrations re-run idempotently on every launch.
• Portable Node 22 runtime — serves the dashboard; nothing to install system-wide.

Everyday use

• A splash screen tracks startup: local database → migrations → gateway → dashboard.
• The gateway and dashboard are supervised with automatic crash recovery.
• Closing the window shuts everything down gracefully — in-flight requests drain and the database stops cleanly.
• A tray icon gives you Open, Restart services, Open logs folder and Quit.
• Per-process logs are written to logs\ (gateway.log, dashboard.log, migrate.log, shell.log, install.log).
• Secrets and ports are generated fresh into native.env (owner-only file ACL) and config.yaml on free loopback ports; an existing configuration is preserved on reinstall.
• Install Ollama locally and the bundled local-ollama provider slot lights up automatically at http://127.0.0.1:11434.

First run

There are no default credentials. On first launch you create the single console administrator account; the password is stored as an Argon2id hash and the setup screen closes permanently afterwards. From there, add a provider in Settings and point any OpenAI-compatible client at the local gateway.

Notes

• Windows x64. The executables are currently unsigned — SmartScreen may warn on first run; choose "Run anyway".
• On the very first launch the local database is downloaded (~60 MB), so the splash screen may sit on "Local database" for a moment.
• Uninstall via Add/Remove Programs or uninstall.ps1; your data is kept unless you pass -PurgeData.

Built from commit b939b11. Licensed under Apache 2.0. Issues and feedback welcome.