Sniff
Tech Stack
The current implementation stack across CLI, agents, multimodal reasoning, and dashboard infrastructure.
Browser Automation
Playwright
Drives the real browser session - navigation, screenshots, mobile device/network emulation.
Vision Reasoning (Tier 3)
Gemini (Vertex AI)
The only provider with documented multimodal input - owns every decision that needs to see the screenshot.
Text Reasoning (Tier 3)
k2-horizon (ifm.ai)
Text-only reasoning: goal enhancement, planning, persona review, copy rewrites.
Fast Decisions (Tier 2)
Jev (Typesafe AI)
Ultra-fast sanity checks - goal-reached detection, context enrichment, decision critique.
Backend API
FastAPI on Cloud Run
Triggers runs/audits, background-task execution, the shared-secret auth gate.
Database
Supabase (Postgres)
Public-read persistence for runs, audits, and schedules - the real system of record.
Scheduling
Cloud Scheduler
Ticks recurring runs/audits on a fixed cadence - a single external trigger, safe across autoscaled instances.
Dashboard
Next.js + Vercel
Web-first frontend - the primary way to trigger and watch runs, not the CLI.
CLI
Python
Secondary, developer/CI-facing entry point - sniff run/demo/experiment commands.