OpenSIN Chat

Self-hosted AI workspace for political research — chat with documents, search the Bundestag, generate reports.

Quick Start | Features | Screenshots | Architecture | Deployment | Credits

---

OpenSIN Chat is a self-hosted AI platform for political work, research, and knowledge management. Upload your documents (Bundestag papers, press releases, legislation drafts) and the AI answers questions only from your sources, with traceable citations. No hallucinations from thin air, no cloud dependency, zero telemetry.

A sovereign, independent product built by OpenSIN-AI and optimized for the German political sphere. Originally inspired by AnythingLLM, OpenSIN Chat has evolved into a purpose-built system for political research with specialized agents, politician databases, and compliance features.

Quick Start

git clone https://github.com/OpenSIN-AI/OpenSIN-Chat.git
cd OpenSIN-Chat/docker-opensin
cp .env.example .env
docker compose up -d

The container maps host port 43939 to internal port 3001. Open http://localhost:43939 after startup.

Important

Node.js 22+ is required for development and bare-metal builds. The root package.json declares engines.node: ">=22.0.0". If you have an older Node version, install Node 22 first (e.g. via nvm install 22 && nvm use 22) or use the Docker setup above which bundles the correct runtime.

Important

Set FIREWORKS_AI_LLM_BASE_PATH and FIREWORKS_AI_LLM_API_KEY in .env to use Fireworks AI as the LLM provider. The SINator Pool Router URL goes in FIREWORKS_AI_LLM_BASE_PATH.

Note

For full setup instructions, environment variables, and bare-metal deployment, see DEPLOYMENT_GUIDE.md.

Features

Core Features

• Document Chat — PDF, DOCX, TXT, Markdown, web pages, YouTube transcripts
• Vector Databases — LanceDB, Chroma, Pinecone, Qdrant, Milvus, PGVector
• LLM Providers — Fireworks AI (primary, via SINator Pool Router); alternatives: OpenAI, Anthropic, Mistral, DeepSeek, Ollama (local), LM Studio
• AI Agents — automated research, web browsing, PDF creation, code execution
• MCP Compatible — integrate any external tool via Model Context Protocol
• Multi-User — permissions, workspaces, audit logs (Docker edition)
• Multilingual — German, English, and more
• Zero Telemetry — no PostHog, no CDN tracking, no outbound calls to third parties

Political Research & OpenSIN-AI Specializations

• Politician Database — Bundestag API + Abgeordnetenwatch as structured sources (biographical data, mandates, votes, speeches). Semantic full-text search over plenary protocols via LanceDB vector index
• Deep Research Pipeline — automated web research (Search → Extract → Summarize) with source tracking. Async via job IDs, polling-capable
• OpenSIN PDF Reports — branded reports (cover, header, footer in OpenSIN blue #009ee0) with table of contents, source lists, and politician references — generated directly from research jobs
• Agent Plugins — @politician-search, @deep-research, @generate-report, @orchestrator, @pdf-analyze, @browser-vision, @image-generation, @create-files — callable directly in chat
• Fireworks AI Vision — multimodal image analysis via Fireworks AI models (minimax-m3, kimi-k2p5/6/7, qwen-3p7-plus). Upload images and the AI describes what it sees
• 3,000+ Tests — comprehensive frontend (Vitest) and server (Jest) test coverage

UI Features

• ChatGPT-style UI — centered chat layout (max-w-800px), user message bubbles right-aligned, AI messages left-aligned
• Dark/Light Mode — toggle button in sidebar, full theme support with light: CSS prefix system
• Code Blocks — syntax highlighting with copy button, language label, bg/border styling in both modes
• Notepad — inline workspace notes with auto-save, pin, and delete (workspace_notes table)
• Grounding Badge — Sparkle icon badge showing when AI uses document sources (RAG)
• Auto-Summary Cards — document snippets preview in sidebar
• Mobile Responsive — 375px viewport, overlay panels, no horizontal overflow
• Loading Animation — 3-pulse dots during AI response
• Action Buttons — hover-only (TTS, Copy, Edit, Good Response, More) like ChatGPT/Claude
• TTS Providers (7 Engines) — Native (browser), OpenAI-compatible, ElevenLabs, Kokoro, Piper (local), NVIDIA NIM, cvoice.ai (German celebrity voices incl. Gronkh, Dieter Bohlen, Joko, Julien Bam, Bushido, Daniela Katzenberger)
• SOTA Charts (ECharts) — the AI generates interactive charts directly in chat: bar, line, area, pie, radar, scatter, treemap, funnel, radialBar. Apache ECharts 6 rendering with gradient fills, glow shadows, rounded bars, staggered elastic animations, dark/light theme support, hover tooltips, save-as-image (2x PNG), and data-view toolbox. Multi-series support for grouped/comparison charts. The create-chart agent plugin sends {type, dataset, title, caption} via WebSocket — no external services, no API keys, fully local rendering

Screenshots

Light Mode	Dark Mode	Mobile

ECharts — SOTA Diagramme direkt im Chat

Die KI generiert interaktive Apache ECharts-Diagramme ohne externe Dienste:

Fraktionsverteilung	AfD nach Bundesland	Redeaktivitat

Top Redner	Abstimmungsverhalten	Ausschuss-Prasenz

Architecture

flowchart TB
    User(["Browser<br/>https://sinchat.delqhi.com"])

    subgraph Cloudflare["Cloudflare Edge (DNS + Tunnel)"]
        DNS["DNS Records"]
        Tunnel["Tunnel Broker"]
    end

    subgraph Host["OCI VM (sin-supabase, 92.5.60.87)"]
        Cloudflared["cloudflared<br/>(Tunnel Client)"]
        Express["Express Server<br/>:3001 (bound to 127.0.0.1:38471)"]
        Collector["Collector<br/>:8888<br/>(Document Parsing, OCR)"]
        Frontend["frontend/dist/<br/>(Static Files)"]
    end

    subgraph Data["Persistence (local)"]
        SQLite[("server/storage/<br/>opensin.db<br/>(Prisma + SQLite)")]
        Files[("server/storage/<br/>uploads/, vectors/")]
    end

    subgraph LLM["LLM Backend"]
        Fireworks["Fireworks AI<br/>(SINator Pool Router)"]
    end

    User -->|HTTPS| DNS
    DNS --> Tunnel
    Tunnel <-->|Outbound Tunnel| Cloudflared
    Cloudflared -->|localhost:38471| Express
    Express -->|"/"| Frontend
    Express -->|"/api/*"| Express
    Express -->|LLM calls| Fireworks
    Express <--> SQLite
    Express <--> Files

    classDef cloud fill:#fff4e1,stroke:#f48120,color:#000
    classDef mac fill:#e3f2fd,stroke:#1976d2,color:#000
    classDef data fill:#f3e5f5,stroke:#7b1fa2,color:#000
    classDef llm fill:#e8f5e9,stroke:#388e3c,color:#000
    class DNS,Tunnel cloud
    class Cloudflared,Express,Collector,Frontend mac
    class SQLite,Files data
    class Fireworks llm

Loading

Repo Structure

OpenSIN-Chat/
├── frontend/          Vite + React 18 + TypeScript + Tailwind + i18next
├── server/            Node.js + Express + Prisma + SQLite/Postgres
│   └── utils/
│       ├── politician/    Politician DB (Bundestag + Abgeordnetenwatch)
│       ├── research/      Deep Research Pipeline
│       ├── reports/       PDF Report Generator
│       ├── orchestrator/  Workflow Engine for Agent Plugins
│       └── agents/        Agent Definitions
├── collector/         Python service for document ingestion and OCR
├── docker/            Original Docker setup (legacy)
├── docker-opensin/    OpenSIN-Chat Docker / Compose setup
├── cloud-deployments/ AWS, GCP, Azure, DO, Helm, OpenShift stubs
├── tests/             E2E and integration tests
├── scripts/           Deploy scripts (deploy-production.sh)
└── docs/              Architecture, ADRs, plans, runbooks

Deployment

Live Demo

https://sinchat.delqhi.com — deployed on an OCI VM (sin-supabase) via Cloudflare Tunnel, Fireworks AI (SINator Pool Router) as LLM provider.

Docker Self-Hosting

cd docker-opensin
cp .env.example .env
# Configure: SERVER_PORT, JWT_SECRET, SIG_KEY/SIG_SALT, LLM keys
docker compose up -d

Bare Metal / Development

See BARE_METAL.md and DEPLOYMENT_GUIDE.md.

Auto-Deploy

An auto-deploy script polls origin/main and rebuilds automatically. Setup in docs/AUTO-DEPLOY.md.

Security Notes

• No credentials in the bundle or repo. Demo/onboarding passwords must never land in the frontend bundle, README, or commits
• Secret rotation. All keys in .env (LLM providers, JWT_SECRET, SIG_KEY/SIG_SALT) are deployment-specific (openssl rand -base64 32)
• Research SSRF protection. The Deep Research Pipeline blocks private/internal targets by default
• Job limits. RESEARCH_MAX_ACTIVE_JOBS (default 3) and ORCHESTRATOR_MAX_ACTIVE_WORKFLOWS (default 2) limit concurrent pipelines

See SECURITY.md for details.

Documentation

• In-app docs: Available at /docs in the running frontend (user manual, API reference, architecture, deployment runbooks)
• Source docs: All Markdown files in docs/ are the single source of truth
• Architecture decisions: ADRs in docs/adr/
• Data sources: docs/DATA-SOURCES.md — external API specs, rate limits, schema mapping
• API reference: docs/api.md

Contributing

1. Fork the repository
2. Create your branch (git checkout -b feature/amazing-feature)
3. Test your changes (yarn test + yarn test:server)
4. Commit and push
5. Open a Pull Request

See CONTRIBUTING.md for details. Code conventions, branching strategy, and commit format are documented there.

License

Distributed under the MIT License. See LICENSE for details.

Credits

OpenSIN Chat was inspired by AnythingLLM by Mintplex Labs Inc. (MIT license), but has since diverged into a fully independent product with nearly 100% of the codebase rewritten or replaced.

We gratefully acknowledge Timothy Carambat and the entire Mintplex team. Their early architectural work provided the initial spark for this project.

AnythingLLM is a full-stack application that enables you to turn any document, resource, or piece of content into context that any LLM can use as reference during chatting. Built and maintained by Mintplex Labs Inc.

What was originally inspired by AnythingLLM: the basic full-stack structure (frontend + server + collector), LLM/embedding/vector DB provider abstraction, and the agent framework concept.

What OpenSIN Chat has built independently (nearly 100% rewritten): complete rebranding (OpenSIN blue, German language, custom logo), all telemetry removed, GDPR-affine defaults, political-use-case specialisation, Politician Database, Deep Research Pipeline, PDF Reports, Agent Plugins, NVIDIA NIM Vision OCR, async upload pipeline with parseJobs, ECharts visualisation, REST APIs, comprehensive test & CI infrastructure, and significant architectural refactoring across frontend, server, and collector.

A full list of third-party components is in THIRD_PARTY.md.

---