tagnos 0.1.0a3

AI-native, multi-judge evaluation and annotation. Your experts, amplified.

Tagnos

AI-native annotation and evaluation where your expertise scales.

Structure your AI evaluation. Capture your experts' judgment. Bring your own LLMs.

Tagnos is the first AI-native annotation and evaluation platform — every action is triggerable by chat, every screen has a context-aware Copilot, and your experts' corrections compound into an AI that speaks with their voice. Built for teams who need traceable, structured AI quality — from solo founders to regulated enterprises.

Powered by CopilotKit · LLM-agnostic via LiteLLM · Apache 2.0

---

Why Tagnos is different

	Feature	What it means
🤖	AI-Native from day 1	Every screen ships with a context-aware Copilot (powered by CopilotKit). Every action is triggerable by natural language. No menu hunting.
🧑‍⚖️	Multi-judge consensus	3 LLMs evaluate each item. When they agree, auto-accept. When they disagree, a human arbitrates. No more single-judge blind spots.
🧠	Expert-Amplifier	Every expert correction is captured into a Knowledge Base that becomes an AI assistant. Junior teammates consult your senior's judgment — even when the senior is offline.
📄🤖	Document + Trace paradigm	Evaluate legal contracts, medical cases, or AI agent traces in the same tool, with the same collaborative workflow.
🔌	Bring Your Own LLMs	Built-in LiteLLM proxy. Configure OpenAI, Anthropic, Gemini, Mistral, or your own Ollama/vLLM — in the UI, no code.
🇪🇺	EU-native, self-host-ready	GDPR by design. Fully air-gapped deployment supported.

---

Quick start

pip install tagnos
tagnos serve

Opens http://localhost:8080 with demo data. Configure your LLM providers in Settings → LLM Providers and start annotating.

Prefer Docker?

docker run -p 8080:8080 tagnos/tagnos

---

Who uses Tagnos

• Legal teams review contracts with AI-assisted clause detection and risk scoring
• Healthcare clinics annotate cases for medical AI training, preserving doctor judgment
• AI product teams evaluate agent traces, catch hallucinations missed by single-judge eval
• Solo AI builders get structured eval without hiring a data science team

---

Comparison

Feature	Tagnos	LangSmith	Phoenix	Label Studio
AI-Native (chat-triggerable actions)	✅ CopilotKit	❌	❌	❌
Multi-judge eval (3 LLMs)	✅	❌	❌	❌
Expert-Amplifier (AI trained on corrections)	✅	❌	❌	❌
Document + Trace in one tool	✅	Traces only	Traces only	Docs only
Bring Your Own LLMs	✅	❌ cloud only	✅	❌
Self-host free	✅	❌	✅	✅
EU-native option	✅	❌	⚠	✅

---

The 5 things Tagnos stands for

1. Your AI learns at the speed of your judgment — not your dev backlog.
2. Your knowledge shouldn't disappear into a spreadsheet.
3. Three LLMs give a second opinion — even when you're alone.
4. The next teammate inherits your reasoning, not a CSV.
5. Every decision today works for tomorrow — without effort.

---

Documentation

• Quickstart
• LLM Providers (BYO-LLM setup)
• Multi-judge explained
• Architecture
• Roadmap

Community

• GitHub Discussions
• Discord (coming soon)

License

Apache 2.0. Commercial use, modification, and distribution permitted.

Contributing

See CONTRIBUTING.md. PRs welcome, especially integrations and plugin examples.