Open-source AI explanations without loading Vestige Index.
UTXO AI is split into two engines: Score AI calculates deterministic risk, while Answer AI explains, guides setup and answers user questions. The extension tries local open-source models first, then calls a separate Cloudflare Worker only when a user asks for help.
The hosted bridge can run Cloudflare Workers AI with Llama when the binding is available. If quota, region or model access is unavailable, UTXO Guard still returns a safe read-only explanation.
Deterministic risk engine
UTXO Score AI
Calculates structured riskPercent from known domain baselines, page signals, wallet prompts, route metadata and L1-L7 checks. It does not chat and does not invent scores.
Conversation and setup engine
UTXO Answer AI
Explains results, helps users install Ollama/Llama, answers Vestige Index questions and can use live web context only when the user allows it. It never changes Score AI output.
Cloudflare Workers AI
Hosted Llama bridge for users who do not want to install a local model.
Ollama
Recommended local runtime. Install, pull llama3.1, and UTXO Guard will detect it automatically.
LM Studio
Desktop local model runtime with OpenAI-compatible local server support.
llama.cpp
Lightweight open-source inference runtime for advanced local deployments.
Recommended command path
Local Llama in three steps.
# 1. Install Ollama from https://ollama.com/download
# 2. Pull a local model
ollama pull llama3.1
# 3. Keep the local endpoint running
ollama serveUTXO Guard checks 127.0.0.1:11434 first. If it cannot reach a local model, it calls the Worker bridge. If the bridge is unavailable, it returns a safe fallback explanation instead of failing silently.
Guardrails