From public documents to accountable actions
CivicFlow-TH explores how AI can use public-service documents to support actions that remain grounded in evidence, reviewable and constrained by policy.
The research architecture combines OpenThai 2.0, developed by iApp Technology with AIEAT, with retrieval-augmented generation, policy checks, evidence and safety gates, and tool orchestration.
Research results and current scope
The team reports 0% hallucinated actions, unsupported actions and unsafe execution across 460 controlled benchmark cases for the full research system. Controlled benchmark results do not guarantee real-world performance.
The current public demo offers guarded retrieval and deterministic policy. Tool execution remains in a sandbox and does not make transactions with government agencies.
Check the public sandbox
This example checks the public health endpoint. Use sample information only when trying the sandbox.
const response = await fetch('https://civicflow-th-demo.vercel.app/health');
if (!response.ok) throw new Error(`HTTP ${response.status}`);
console.log(await response.json());