Set & Signal 🏋️ · Make the next set obvious

Set & Signal is a self-hostable workout planner and training log for planning sessions, recording sets, reviewing history, and asking an AI assistant about training data. The product ships as a React/Vite web app embedded in a Go server, with passkeys/OIDC sign-in, Convex-backed training records, SQLite for identity and operational state, and an MCP endpoint for assistant clients.

A session in the real world: plan a push day, log the working sets as you go, then ask the assistant what to train next. Your history, prescription, and account boundary stay in the same self-hosted system.

The project is designed for a single deployment you control. It includes an exercise catalogue, progression and prescription helpers, workout history, bodyweight tracking, reminders, push notifications, localization, and a read/write MCP surface protected by the same account boundary as the web app.

Run locally

Requires Go 1.27+, Node.js 24+, pnpm 12+, PostgreSQL-compatible Convex deployment, and an OAuth or passkey configuration. The API refuses to start without CONVEX_URL; the full training setup and storage rules are in docs/convex.md.

Start the API and web app together, then open https://set-and-signal.localhost:

pnpm install --frozen-lockfile
pnpm dev

pnpm dev runs dev:api and dev:web in parallel. The API reads CONVEX_URL from web/.env.local and stores data in ./data.

Both run through Portless (a dev dependency): the API at https://api.set-and-signal.localhost, the web app at https://set-and-signal.localhost. The first run may ask for sudo to bind port 443 and trust a local certificate. Vite proxies /api, /oauth, /.well-known, and /mcp to the API, so OAuth callbacks, session cookies, and the local MCP endpoint all use the web origin. Passkeys registered for localhost do not carry over to the new hostname.

For a fresh development deployment, run pnpm exec convex dev from web and follow the setup guide to configure AUTH_ISSUER and AUTH_JWKS. Keep the generated web/.env.local, signing keys, and any provider credentials out of Git.

Related MCP server: tally-workout-mcp

Verify

go test -race ./...
go vet ./...
(cd .railway && go test ./... && go vet ./...)

The service is licensed under the GNU AGPL v3 or later. See SECURITY.md for vulnerability reports.

Architecture

flowchart LR
  Browser[React/Vite web app] --> API[Go HTTP API]
  Browser --> Convex[Convex training functions]
  API --> SQLite[(SQLite identity + operations)]
  API --> Convex
  API --> MCP[MCP Streamable HTTP]
  API --> Auth[Passkeys / OIDC]
  API --> Push[Web Push + reminders]

The Go composition root is cmd/opengym-api. HTTP and MCP protocol concerns live in internal/httpapi; training state, analytics, scheduling, and progression live in internal/training; SQLite access and migrations live in internal/store. The frontend follows the app → features → domain → shared direction described in docs/architecture.md. Edit exercise content in web/catalog; generated instruction files are rebuilt by the web scripts.

Configuration essentials

.env.example documents the complete runtime surface. The settings most people need are:

Further documentation

• Architecture — dependency direction and build layout.

• Convex setup and storage — hosted training ownership, offline queues, keys, and current format requirements.

• Data model — storage ownership and current contracts.

• Product and design — user flows and UI constraints.

• Railway deployment — infrastructure configuration.

Status and privacy

The repository is an active, self-hosted application with hosted Convex training storage. Training records are account-scoped, but a deployment's operator controls the API, Convex project, and persistent volume. Review SECURITY.md and configure access controls before exposing an instance to other users.