Fidechat monorepo: a Discord clone with real-time messaging. It was created by merging two independent repositories (now archived), keeping the full commit history of both:
| Workspace | Package | Origin (archived) | Stack |
|---|---|---|---|
apps/api |
@fidechat/api |
Datzu712/Fidechat-backend | NestJS 12 · Fastify · PostgreSQL + Drizzle · Keycloak · Socket.IO |
apps/web |
@fidechat/web |
Datzu712/fidechat-web | Vite · React 19 · Tailwind v4 · Radix UI |
Fidechat started as a university project and has gone through three versions. The first two were written without AI assistants; the third is an AI-assisted refactor.
- v1 and v2 were university projects, each built within a single four-month term (cuatrimestre). They were written against a deadline, so the code is rushed in places: they are kept here as a record of how the project grew, not as examples of best practice.
- v3 is the final version, and the peak of it all: the project I started at university, rebuilt without a deadline, done properly this time, with everything that never fit into a term.
| Version | Dates | Backend | Frontend | Tags | Built with AI |
|---|---|---|---|---|---|
| v1 | Nov 2024 – Feb 2025 | Java 21 · Spring Boot 3.1 · JDBC (PostgreSQL) · JWT · WebSockets | React 18 · Vite | api-v1-java · web-v1-vite |
No |
| v2 | Jul 2025 – Aug 2025 | NestJS 11 · Fastify · Oracle (PL/SQL packages and triggers) · Keycloak · Socket.IO | Next.js 15 · shadcn/ui · NextAuth | api-v2-nest-oracle · web-v2-next |
No |
| v3 | Aug 2026 – today | NestJS 12 · PostgreSQL + Drizzle · Turborepo monorepo | Vite SPA · Tailwind v4 | — | Yes, with Claude |
How to verify it:
- The original repositories are archived (read-only) on GitHub, with their original hashes, pull requests and Dependabot PRs: Fidechat-backend and fidechat-web.
- Merging them here moved the files into
apps/*, which changes every hash. docs/history-map.md maps each monorepo commit to its original commit on GitHub (171 of 171). - Every v3 commit, and only those, carries a
Co-Authored-By: Claude …line. To list them:git log --grep='Co-Authored-By: Claude'. - The OpenAI chat bots present since v2 are a product feature, not AI-generated code.
Running an old version: legacy/ has one-command Docker Compose environments that build the exact code of each tag, with demo data:
docker compose -f legacy/v1/compose.yaml up --build # v1 -> http://localhost:4100
docker compose -f legacy/v2/compose.yaml up --build # v2 -> http://localhost:4200Warning
v1 and v2 are old and full of known vulnerabilities: outdated dependencies, missing authorization checks and more. They are here to be looked at, never to be deployed. Run them only locally, and do not use them in production or expose them to a network.
Screenshots of each version are in docs/screenshots/.
fidechat/
├── apps/
│ ├── api/ # NestJS backend (@fidechat/api)
│ └── web/ # Vite SPA frontend (@fidechat/web)
├── packages/
│ └── shared/ # contracts shared by both apps (@fidechat/shared)
├── legacy/ # reproducible environments for v1 and v2
├── docker/ # demo realm and seed for compose.yaml
├── docs/ # history-map.md (hashes ↔ original repos), screenshots/
├── .github/ # CI workflows
├── compose.yaml # quick launch of v3 (see below)
├── package.json # workspace root
├── pnpm-workspace.yaml
└── turbo.json
- Node.js >= 26.13 (current LTS)
- pnpm >= 12
- Docker + Docker Compose (for Postgres and Keycloak)
pnpm install # install dependencies for every workspace
pnpm dev # run api and web together
pnpm build # build everything
pnpm lint # lint everything
pnpm test # test everythingUse --filter to work on a single workspace:
pnpm --filter @fidechat/api dev
pnpm --filter @fidechat/web dev
pnpm --filter @fidechat/api add <package>To try the current version without setting anything up, compose.yaml builds
the api image from your checkout (it also serves the web SPA) and runs it with
PostgreSQL, Keycloak and demo data. It needs only Docker with Compose:
docker compose up --build # -> http://localhost:4300Log in as ana or bruno (or their @demo.local emails), password
demo1234. Keycloak's console is at http://keycloak.localhost:4380 (admin /
admin). Stop with docker compose down; add -v to wipe the database.
The ports are fixed, because the Keycloak realm's redirect URIs use them, and
the demo users and data live in docker/.
Note
AI bots. They are optional: without a key they show up online but stay
silent. To turn them on, set ANTHROPIC_API_KEY in the shell or in a root
.env (git-ignored, so it is the right place for it). The model defaults to
anthropic:claude-haiku-5-5; set AI_MODEL=openai:gpt-6-luna (the
recommended OpenAI model) and OPENAI_API_KEY to use OpenAI instead, or
AI_MODEL=ollama:<model> (no key) to use an Ollama running on the host.
The bots don't answer every message. After a human writes, an LLM router reads
the channel and decides who, if anyone, should speak, like a person in a group
chat would (BotConversationService in apps/api).
flowchart TD
A[Human message] --> B[Wait 0.6 s of channel silence]
B --> C{Router reads the last 20 messages}
C -->|nobody| Z[No reply]
C --> D[Bots spoken to reply in parallel<br/>a named bot always answers]
D --> E[A bot joining on its own replies next,<br/>after reading the others]
E --> F[At most one other bot may react]
- Replies are short chat messages, posted after a typing delay that matches their length.
- A bot joins on its own at most once every 10 minutes per channel; this cap lives in code, not in the prompt.
- A new human message cuts the cycle short: bots nobody named stop, named bots still answer, and the router decides again.
- With
AI_ROUTER=off, or if the router fails, only bots named in the message answer. - With the router on, every human message sends the last 20 messages of its channel to the model provider, even when no bot is named. Fine for a demo; a real deployment would want a per-guild opt-in.
The compose files live in apps/api and are still run from there:
cd apps/api
cp .env.sample .env
docker compose -f local-docker-compose.yaml up -d # Postgres + Keycloak
pnpm db:migrate # create the schema with DrizzleThe backend production build (apps/api/dockerfile.prod) uses the monorepo
root as its context, because there is now a single pnpm lockfile, at the root. It also builds the web
SPA and serves it from the API, so the browser and the API share one origin.