@tidegap/core

Renamed from @ebb-ai/core in 0.18.0. See Renamed from Ebb AI.

Workload scheduling for the agentic-AI economy — the core TypeScript library.

@tidegap/core lets your code defer non-urgent LLM tasks to cheap, off-peak grid windows. Anthropic and OpenAI Batch APIs already give a flat 50% discount (projected, provider list price) for 24h-tolerant work; @tidegap/core makes the choice automatic, picks the cheapest window inside your deadline, and writes an auditable receipt (cost + carbon + provider + duration) for every dispatch.

It ships:

• defer(fn, { deadline, region, carbonBudgetG }) — wrap any async task.
• Scheduler — opt-in SQLite-backed durable queue (new Scheduler({ dbPath })).
• Provider adapters for Anthropic + OpenAI, with sync and Batch API paths.
• Grid carbon-intensity feeds:
  • mockGridFeed() — deterministic synthetic curve.
  • electricityMapsFeed(apiKey) — Electricity Maps free-tier API.
  • ukCarbonIntensityFeed() — UK National Grid ESO (free, no key, GB only).
  • multiSourceGridFeed({ feeds, fallback }) — per-zone routing.
• recommendWindow({ region, deadline }) — planning-only API; returns the cleanest window without queueing.

Install

npm install @tidegap/core
# optional adapters; install only what you use
npm install @anthropic-ai/sdk openai

Quick start

import { defer } from "@tidegap/core";

const summary = await defer(
  async () => callAnthropic("Summarize today's CHANGELOG"),
  { deadline: new Date(Date.now() + 12 * 3600 * 1000).toISOString(), region: "US-CAL-CISO" },
);

Documentation

Full docs, MCP server, dashboard, and architecture in the monorepo README.

License

Apache-2.0.