Showcases
Real workflows, real teams, real savings. These are production deployments and community projects that replaced fragile prompt loops with compile-once, deterministic AINL graphs.
7.2×
Cost reduction (internal cron fleet)
0
Orchestration LLM calls at runtime
100%
Deterministic execution
Rebuilding a routine monitoring agent with AINL: compile-once orchestration, no runtime LLM loops, strict validation, JSONL audit tapes, and OpenClaw cron + Hermes-friendly emits — with real files and the published cost savings report.
A 30-line AINL workflow checks your inbox every 15 minutes, skips empty polls silently, and fires a Telegram notification the moment unread email arrives — no LLM tokens, no fragile Python script, no missed messages.
A compiled AINL graph polls four services every 5 minutes, auto-restarts downed processes with cooldown gating, persists a 7-day restart history, and emits structured health envelopes to your alerting queue — deterministically, with a full JSONL execution tape.
Stop chasing overdue invoices manually. A compiled AINL workflow queries your DB daily, identifies invoices past 30 days unpaid, totals the exposure, and emails a digest — deterministically, no LLM required at runtime.
Run a nightly AINL workflow that scores every lead in your CRM for data completeness, flags quality drops above a threshold, and alerts your team — without touching a line of Python or paying LLM tokens to decide what to audit.
Replace a chatty, token-burning morning assistant with a compiled AINL workflow that checks calendar, email, and context — then delivers a briefing at 9 AM without a single orchestration LLM call.
Track product prices or financial symbols with a compiled AINL scraper that runs hourly, persists to Postgres, and alerts on threshold crossings — without an LLM deciding whether to check prices each time.
Track daily and weekly LLM spend across your team with a compiled AINL workflow — hourly OpenRouter usage pulls, rolling 7-day budget calculations, threshold alerts, and a structured audit envelope. All without an LLM deciding whether to check your LLM bill.
How one developer replaced a fragile Tweepy + LangChain loop with a single compiled AINL graph that searches, classifies, replies, deduplicates, and respects rate limits — without burning tokens on every poll.
Drop a description in GitHub Discussions or email us directly. We'll help you write it up and feature it here. We're especially interested in:
